AI Distilled

AI agents power up Opera’s browser, Colab; Gitingest provides text digest of codebases, Deepseek relAI_Distilled #85: Your AI News Fix!Protect Data Privacy and Optimize AI Models with Tonic TextualLLMs have tapped all of pubically available data. The last mile training of models requires private data. Use private data without compromising security. Redact, label, and prep freetext for LLM ingestion or data pipelines.START FREE TRIALIt looks like the AI giants are battling it out, with announcements on new models, Gen-AI capabilities for their flagship products, and research breakthroughs. But don’t you worry, we’ve got you. Here is your weekly digest!LLM Expert Insights Team,Packt📰 NewsThe 2024 ACM A.M. Turing Award goes to Andrew G. Barto and Richard S. SuttonKnown for their pioneering research in reinforcement learning Barto and Sutton’s decades long research has shaped AI agents, robotics, and gaming. The 2024 Turing Award recognizes their profound contribution to AI and ML.AI steals the thunder at MWC 20251. Deutsche Telekom’s AI phone Deutsche’s upcoming AI phone, equipped with an AI assistant powered by Perplexity, will be available to the public later this year.2. OPPO Announces Enhanced AI Strategy OPPO has announced its AI strategy, featuring innovations like AI Call Translator and AI VoiceScribe to level up their mobile AI experiences.3. Stability AI and Arm Bring On-Device Generative Audio to SmartphonesStability AI and Arm’s partnership is set to enable high-quality sound effects and audio sample generation directly on mobile devices, making it 30x faster on Arm CPUs.4. Google Showcases Android’s AI and Gemini Features; Wins Two GLOMO Awards at MWC 2025Google demoed Android AI Core, featuring smart replies and text summarization, powered by Gemini Nano. Google’s Gemini won the Breakthrough Device Innovation highlighting Google’s leadership in AI for mobile. Pixel Pro, additionally, was named Smartphone of the Year.Google Switches on AI Mode in LabsGoogle is testing AI Mode in Labs, an experimental search experience, for its Google One AI Premium subscribers. Powered by Gemini 2.0, AI Mode expands on AI Overviews offering more advanced reasoning, thinking and multimodal capabilities.Google’s March Pixel Drop Gemini Live will support multilingual conversations in 45+ languages and expand iPixel’s multimodal capabilities, with support from Gemini Nano for On-device AI.World’s 1st Commercial Biological Computer Launched by Australian Start-UpCortical Labs, an Australian startup, introduced CL1, the world's first commercial biological computer, at MWC. This "body in a box" uses living human brain cells to grow neurons capable of learning and processing information biologically, consuming far less energy than traditional AI. This “Wetware-as-a-Service” computer is set to launch in the second half of 2025.Google Releases Teaser for Gemini’s AI-Powered Video AnalysisNow, Gemini can analyze live videos with its vision capabilities. Users can share their screen or stream videos directly from their device camera to receive real-time insights from Gemini. This update is expected to roll out for Google One AI Premium users later this month.Opera Previews Its Agentic AI Browser OperatorOpera is testing an AI agent integrated into its browser. With this native AI agent, Opera aims to offer efficiency and user control while assisting with browsing tasks.AI Jam Session Anthropic has closed a Series E funding round, bringing its post-money valuation to $61.5 billion. This funding will support Anthropic’s expansion plans and the development of next-generation AI technology.To create a culture of transparency and trust in AI, Anthropic also launched the Transparency Hub to provide information about its AI models, safety research, model evaluations, and methodologies.Apptronik and Jabil Partner to Scale Apollo Humanoid RobotsApptronik and Jabil have teamed up to build and integrate humanoid robots for tasks like inspection, sorting, and delivery.Sanctuary AI Integrates Sensors into Phoenix RobotsSanctuary AI is equipping its Phoenix humanoid robots with tactile sensors to enhance dexterity and precision in handling delicate tasks. This upgrade will improve Phoenix’s manipulation capabilities for real-world applications by introducing a sense of touch.Figure Accelerates Helix’s Launch Timeline by Two YearsCEO Bret Adcock announced that Helix will enter Alpha testing this year, with the humanoid expected to reach households earlier than anticipated.Amazon's Ocelot Chip Advances Practical Quantum ComputerThe AWS Center for Quantum Computing has introduced Ocelot, a new quantum computing chip designed to make quantum computing more feasible. The Ocelot prototype aims to reduce the cost of quantum error correction by up to 90% compared to existing methods.💻 Awesome AI: Tools for WorkAlibaba’s Open Weight QWQ-32B Reasoning ModelAlibaba has released QWEN-32B that uses reinforcement learning. Designed to be highly performant, QWQ-32B reports results comparable to much larger models.Data Science Agent in Google Colab, Powered by GeminiGoogle has now released its new AI agent for Colab in select countries and languages. Designed for users 18 and older, this Data Science Agent simplifies data analysis by automating Jupyter notebook creation from text prompts. It can handle tasks like data loading, library imports, exploratory analysis, and visualization code generation.Cohere’s Open-Source Aya Vision Model for Multilingual and Multimodal UnderstandingCohere AI has introduced Aya Vision, a state-of-the-art vision model designed to bridge language gaps in AI, especially for multimodal tasks combining text and images. Aya Vision can perform image captioning, visual question answering, and text generation across 23 languages. Available in 8B and 32B parameter sizes, the model is accessible via open-source platforms and WhatsApp for research and non-commercial use.Convert Your Git Repos into Text with GitingestGitingest is an open-source tool that converts Git repositories into text for LLMs. It simplifies code analysis and AI solutions by providing a structured, prompt-friendly text digest of codebases. Features include smart formatting, statistics on file structure, and CLI/Python package usage.Flow Releases Integrations for Popular AI AppsWispr Flow is an AI voice dictation tool that uses real-time voice-to-text conversion to allow users to type up to three times faster. It features AI commands, auto-editing, and supports over 100 languages. Context-aware and adaptable to individual speech patterns, it caters to professionals, writers, and students, with tiered pricing options.Google’s Confidential Federated AnalyticsGoogle Research has introduced Confidential Federated Analytics (CFA), a privacy-preserving technique that prioritizes user privacy while discovering new words to improve search engines. CFA analyzes anonymized and aggregated search query data from numerous devices, without inspecting individual queries directly. This technique helps identify emerging words and trends, improving search quality, particularly for low-resource languages.ATLA AI Releases Frontier LLM Evaluation Model Selene-1 Selene-1 is a powerful LLM evaluator model equipped with absolute scoring, classification, and pairwise preference capabilities. With customizable evaluations and chain-of-thought critiques, Selene-1 can detect hallucinations and verify the accuracy of LLM responses.Create Natural and Intuitive HCI Through Speech and LanguageRecently launched Sesame AI employs a Conversational Speech Model (CSM) to create human-computer interaction interfaces using speech and natural language.🛠️ HackhubConvert Your Git Repos into Text with GitingestGitingest is an open-source tool that converts Git repositories into text for LLMs. It simplifies code analysis and AI solutions by providing a structured, prompt-friendly text digest of codebases. Features include smart formatting, statistics on file structure, and CLI/Python package usage.Flow Releases Integrations for Popular AI AppsWispr Flow is an AI voice dictation tool that uses real-time voice-to-text conversion to allow users to type up to three times faster. It features AI commands, auto-editing, and supports over 100 languages. Context-aware and adaptable to individual speech patterns, it caters to professionals, writers, and students, with tiered pricing options.Google’s Confidential Federated AnalyticsGoogle Research has introduced Confidential Federated Analytics (CFA), a privacy-preserving technique that prioritizes user privacy while discovering new words to improve search engines. CFA analyzes anonymized and aggregated search query data from numerous devices, without inspecting individual queries directly. This technique helps identify emerging words and trends, improving search quality, particularly for low-resource languages.Create Natural and Intuitive HCI Through Speech and LanguageRecently launched Sesame AI employs a Conversational Speech Model (CSM) to create human-computer interaction interfaces using speech and natural language.⚙️TechhubOpenAI’s NextGenAI Research and Education ConsortiumOpenAI has launched NextGenAI, a consortium of 15 research institutions, backed by $50 million in grants, compute funding, and API access. This initiative supports students, educators, and researchers in pushing the boundaries of AI knowledge and preparing future AI leaders. Founding partners include Caltech, Duke, Harvard, MIT, Oxford, and more, alongside institutions like Boston Children's Hospital and the Boston Public Library.DeepSeek Releases SmallPond for Distributed Data ProcessingDeepSeek AI has introduced SmallPond, a lightweight data processing framework designed for high-performance AI training and inference on large datasets. Built on DuckDB and DeepSeek's 3FS, it efficiently processes petabytes of data using distributed processing with Ray.📖 New Title ReleasesBUY NOWBUY NOWBUY NOW📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

DeepSeek open sources 5 repos for AGI, Helix and Engine AI’s humanoids gain more power, Agents in acAI_Distilled #84: Your AI News Fix!You can now train your own Reasoning model like DeepSeek-R1 locally with just 5GB VRAM. Unsloth is fully open-source and allows you to transform any open LLM like Llama 3.1 (8B) or Phi-4 (14B) into a reasoning model.GitHub repo: https://github.com/unslothai/unslothDeepSeek’s R1 research revealed an “aha moment” where R1-Zero autonomously learned to allocate more thinking time without human feedback by using Group Relative Policy Optimization (GRPO). Unsloth enhanced the entire GRPO process, making it use 90% less VRAM than all other implementations. This allows you to reproduce R1-Zero's "aha moment" on just 5GB of VRAM using Qwen2.5 (1.5B).Try Unsloth's free GRPO notebook with a free 16GB GPU: Llama 3.1 (8B) on ColabFor a Tutorial and GRPO notebooks featuring other models like Phi-4, visit Unsloth's docsIt looks like the AI giants are battling it out, with announcements on new models, Gen-AI capabilities for their flagship products, and research breakthroughs. But don’t you worry, we’ve got you. Here is your weekly digest!LLM Expert Insights Team,Packt📰 NewsDeepSeek open sources five repos for AGI in its OpenSourceWeekIn its OpenSource week, DeepSeek is making available five repos that form the building blocks of their online service. These repos include FlashMLA (efficient MLA decoding kernel for Hooper GPUs), DeepEP (EP communication library for MoE model training and inference), DeepGEMM (FP8 library supporting dense and MoE GEMMs), and DualPipe (a bidirectional parallelism algorithm), and Fire-Flyer File System (a parallel file system that utilizes the full bandwidth of modern SSDs and RDMA networks).Microsoft’s next generation of Phi-4 modelMicrosoft introduced Phi-4-multimodal and Phi-4-mini, the latest additions to Microsoft's Phi family of small language models (SLMs). Phi-4-multimodal handles speech, vision, and text concurrently, while Phi-4-mini is proficient in text-based tasks. Phi-4-multimodal is a 5.6B parameter model, and Phi-4-mini is a 3.8B parameter model. Both models are suitable for compute-constrained inference environments.Google announces public preview of Gemini Code AssistGoogle has made Gemini Code Assist available to individual developers for free public preview, with a liberal token window of 128K. This AI-coding assistant offers code completion, generation, and chat features in Visual Studio Code and JetBrains IDEs, similar to thosealready available in Firebase and Android Studio. And guess what, you have about 180,000 code completions every month! Insane! isn’t it? A similar tool, Gemini Code Assist for GitHub, is also available, providing AI-powered code reviews.Amazon introduces Gen-AI infused Alexa – Alexa+Amazon introduced Gen-AI-powered Alexa+ this week. It features agentic capabilities and is designed to be smarter than the original Alexa, with LLMs powering up its knowledge base. Designed to take actions, it can remember your specific needs and requirements, making your experiences more useful and personalized. Available on Echo devices, a new mobile app, and Alexa.com, it costs $19.99 per month but is free for Prime members.Claude’s 3.7 Sonnet Hybrid reasoning with extended thinking and Claude CodeAnthropic has announced Claude 3.7 Sonnet with hybrid reasoning capabilities. Users can now toggle between fast responses and extended thinking modes, with a budget of up to 128K tokens. Unlike other reasoning models, Claude is more focused on the real-world business applications of LLMs, rather than math and computer science competition tasks. Anthropic also introduced Claude Code, a command-line collaborative tool for agentic coding, currently available as a limited research preview.Alibaba’s open-sources thinking model QwQ-Max-PreviewAlibaba, through an announcement blog post created by QWQ-Max-Preview, unveiled the newest model in the Qwen series: QwQ-Max-Preview. It is built upon Qwen2.5-Max and excels in mathematics, coding, general tasks, and agentic workflows. The post also mentions future plans, which include the development of a dedicated app for Qwen Chat and smaller QwQ variants for local device deployment.Comet an agentic search browser by PerplexityPerplexity announced its agentic browser Comet in an X post. Built on the Chromium framework, Comet will integrate search and automate related tasks. It will also integrate deep research and real-time information processing. You can join the waitlist here.Perplexity also announced voice mode for its iOS app. Voice mode is expected to be shipped for Android and Mac apps in the coming days.Microsoft cancelling U.S. data center leases amid CEO Satya Nadella’s concerns about AGI milestonesA TD Cowen report states that Microsoft has pulled the plug on 200MW leases for at least two private data centers, withdrawn from around 500 leases, and reallocated a sizeable portion of its international spend to the US. In another development, CEO Satya Nadella, shared his thoughts on AGI hype. He opined that self-proclamation of AGI is useless and the true revolution, the real benchmark will be when we see growth in the GDP. “It can’t be just supply side,..,when the productivity goes up, and the economy is growing at a faster rate. When that happens… that’s to me is the moment,” he said.Alibaba to invest RMB 280 billion in AI and cloud computing infrastructureAlibaba plans to invest USD 53 billion over the next three years to scale up AI capabilities and cloud infrastructure, providing businesses with tools for innovation. CEO Eddie Wu sees AI as a "once-in-a-generation" opportunity. Cloud computing is Alibaba's main revenue driver in AI, with high demand for AI hosting services. Alibaba is integrating AI across its ecosystem to improve customer experiences, optimize business operations, and drive long-term growth.Apple makes $500 billion commitment to US’s future – Tim Cook, CEO, AppleApple plans to invest over $500 billion in the U.S. in the next four years, focusing on investments in AI, silicon engineering, manufacturing, and skills development. A new manufacturing facility will be opened in Houston for Apple Intelligence servers and the U.S. Advanced Manufacturing Fund will be doubled to $10 billion. A manufacturing academy will be established in Michigan, and R&D investments will expand across the U.S., creating about 20,000 jobs. Apple continues to support educational programs for hardware engineering and silicon chip design.SamA announces two new features for ChatGPT Plus and free usersOpenAI released research preview for GPT 4.5 this week to understand its strenght and limitations.In his X posts, OpenAI CEO, Sam Altman, announced DeepResearch for ChatGPT Plus users and Advanced Voice for GPT-4o mini.In another development, The Information reported that OpenAI plans to shift 75% of its data center capacity to StarGate, financed by SoftBank. This transition from Microsoft-owned data centers is expected to occur over the next five years.Meta for Education, a new mixed and virtual reality (VR/MR) offering, is now generally available. It provides educators with Meta Horizon-managed solutions, aimed at enhancing student engagement and knowledge retention through interactive VR/MR experiences.💻 Awesome AI: Tools for WorkAlibaba releases wan 2.1 family of video modelswan2.1 presents two versions of video generation models: a lightweight 1.3 billion parameter model suitable for laptops, and a robust 14 billion parameter model for higher performance. wan2.1 handles both text-to-video and image-to-video generation, providing resolution choices of 720p or 480p. It can simulate complex motion, capture intricate details, and generate multilingual text effects.Pika announces Pika 2.2, PikaFrames, andPikaswaps on XPikaswaps allows users to modify and replace objects in videos using video inpainting. It enables the swapping, erasing, and altering of objects while maintaining realistic visual consistency. Features include a brush tool, reference image uploads, and options to re-prompt or retry.Engine AI’s humanoid can perform complete front flipEngineAI has unveiled the world's first humanoid robot capable of performing a front flip. This achievement marks a significant advancement in humanoid robotics, showcasing improved agility and control. The robot's ability to execute complex acrobatic movements demonstrates advancements in AI-driven motion planning and real-time control systems.Grok3 voiceIn his X post, CEO, Elon Musk announced that xAI’s Grok3 has enabled conversation mode for Premium and SuperGrok users..Helix – A vision language action modelFigure AI’s Helix model is designed to bring humanoid robots into homes. It blends computer vision, language comprehension, and real-time motor control. Helix can adapt on the go, learn quickly with minimal training data, control multiple robots simultaneously, and handle thousands of household items. It runs on embedded low-power GPUs And can pick up virtually any small household object by voice command. 🛠️ HackhubMagma: A foundation model for multimodal AI agents across digital and physical worlds - Microsoft ResearchMicrosoft Research has introduced Magma, a foundation model for multimodal AI agents, to bridge the digital and physical worlds. Magma integrates diverse sensor data—such as vision, audio, and depth—enabling agents to perceive and interact with complex environments. It supports a wide range of tasks, from simple object recognition to intricate navigation and manipulation. It can create adaptable agents that can learn and generalize across various scenarios, enhancing robotics, AR/VR, and human–computer interaction.Meta’s ML GymMLGym is an open-source framework and benchmark designed to accelerate AI agent research. It aims to simplify the development, evaluation, and comparison of AI agents across diverse environments. By offering a standardized platform for researchers to conduct experiments, share results, and collaborate, MLGym will enable more efficient and reproducible research.PaliGemma 2 - New Instruction Vision Language Models by GooglePaliGemma2-Mix is a vision-language model based on the Gemma language model and SigLIP vision model. Optimized for efficiency and performance, the model is available on Hugging Face. It's designed for tasks requiring visual understanding and language generation, such as image captioning and visual question answering. The "mix" version provides a blend of pre-training and fine-tuning, offering a versatile and robust model.⚙️TechhubGibber link – AI Agent communication protocolGibber Link is an agent communication protocol that proposes the use of sound-level protocols instead of speech for efficient communication. This reduces compute costs by 90%, speeds up data transfer by 80%, and minimizes errors. The protocol automatically switches from speech to sound upon detecting another AI agent, enhancing clarity and enabling multimodal data exchange.Meta MotivoMeta Motivo is a tool by Meta Demolab that can be used for creating 3D character animations from audio inputs. It uses audio-driven motion generation and analyzes speech patterns to produce realistic facial expressions and body movements. Motivo employs a neural network trained on a large dataset of speech and motion capture data, enabling it to synthesize animations that synchronize with the audio.Introducing the SWE-Lancer benchmark | OpenAIOpen AI’s SWE-Lancer is a benchmark of over 1,400 freelance software engineering tasks from Upwork valued at $1 million. It features bug fixes, feature implementations, and managerial tasks graded by experienced engineers. Designed to study the economic impact of AI models, SWE-Lancer offers a unified Docker image and the open-sourced SWE-Lancer Diamond for future research.🧠MasterclassGenerative Ghosts: Anticipating benefits and risks of AI afterlives - Google DeepMindGoogle DeepMind is working on "generative ghosts," AI agents representing deceased individuals, which are becoming increasingly common due to advances in generative AI. The research work explores design of these agents, considering factors like provenance, embodiment, and representee type. This paper also investigates inner AI misalignment, focusing on how training steering signals can cause harmful behaviors. It introduces “evil steering,” where innocuous steering creates aligned-but-malevolent agents, even with proper reward design for helpfulness. Grid world experiments demonstrate that steering during learning can cause negative outcomes despite well-designed rewards. Latent space analysis reveals “evil steering” mechanisms.Findings emphasize carefully considering steering, not just rewards, for AI safety, preventing unintended emergent behaviors.Delta Variances - Google DeepMindGoogle’s recent work introduces Delta Variance, an efficient algorithm for quantifying epistemic uncertainty in neural networks. It addresses the challenge of estimating uncertainty arising from limited data, which is crucial for reliable decision-making. The algorithm requires no modifications to network architecture or training. It offers a unified view of related methods and showcases improved performance through empirical results, including a weather simulation example.Test time scaling -zero risk response – John Hopkins UniversityThis work investigates whether increasing the inference-time compute budget improves model confidence in its answers. Models are evaluated in a selective question answering setting, where they can choose to abstain from answering.The results indicate that with increasing compute budget, the confidence in correct answers improves, but the confidence in incorrect answers decreases. They propose a new evaluation metric, utility, that considers both accuracy and confidence and show that the approach improves performance on Jeopardy Odds and Exam Odds benchmarks.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!👉Tell us more about your content needs We would love to hear from you! Fill out this form to tell us what you’d like to read in AI Distilled next. *{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Perplexity makes DeepResearch free to use, MoonShot AI’s MoBA for long context LLMs, DeepSeek’s Code AI_Distilled #82: We are back! CLICK HERE TO REGISTER Get Early Access - 40% Discount Use Code AGENT40 at checkout This week saw many breakthroughs and announcements and we bring them all together in one place. Our goal is to curate the most relevant news and updates for you. Fill out this form and tell us what you’d like to read next on AI Distilled. LLM Expert Insights Team, Packt 📰 News Google introduces AI co-scientist The AI co-scientist by Google is a multi-agent AI system that is intended to function as a collaborative tool for scientists. It is built on Gemini 2.0 and is designed to mirror the reasoning process underpinning the scientific method. The AI co-scientist can be used to generate novel research hypotheses, a detailed research overview, and experimental protocols. Microsoft unveils Majorana 1 Microsoft has developed a the topoconductor that allows them to create topological qubits and engineer a new state of matter. The toplogical qubits are more stable than traditional qubits, making them more suitable for building a large-scale quantum computer. Microsoft is now gearing towards the next step - building a fault-tolerant quantum computer using topological qubits. Thinking Machines Lab launched, Mira Murati CEO Former OpenAI CTO Mira Murati has launched a startup with Barret Zoph (CTO), John Schulman (Chief Scientist). Various other AI stalwarts who have experience in creating AI products like ChatGPT, Segment Anything, Mistral, Pytorch, Character.ai, OpenAI Gym, and FairSeq are also a part of Thinking Machines Lab. The startup’s core mission is to build intelligent, adaptable, and personalized AI systems, emphasizing human-AI collaboration and safety. It aims to make AI more capable, customizable, understood, and user-friendly. Perplexity Deep Research launched and is free to usePerplexity recently launched its Deep Research model, designed to generate comprehensive reports, using capabilities like iterative search, reasoning, coding, and refinement of research plans. On the Humanity’s Last Exam benchmark test, Perplexity ranked second—behind OpenAI’s deep research model but ahead of other leading competitors—completing most research tasks in under three minutes. Google aims to serve 10 cities with Waymo self-driving cars in 2025 Speaking at the 2025 World Government Summit in Dubai, Google and Alphabet CEO Sundar Pichai talked about expanding Waymo to 10 new cities. He also highlighted Google’s recent achievement in quantum computing and indicated that quantum computers could become mainstream in the next 5 to 10 years. Isomorphic Labs and Novartis expand collaboration Google DeepMind partner Isomorphic Labs, an AI-first drug discovery company, and Novartis have extended their collaboration to add three more research programs aimed at accelerating drug discovery research. Isomorphic Labs is augmenting the AlphaFold breakthrough to connect research with biotech, drug discovery, and medical design. HP acquires Humane’ AI capabilities including the AI platform Cosmos; end of the road for AI Pins HP is acquiring Humane in a $116 million deal to accelerate the development of an intelligent ecosystem across its products and services. Humane has also announced the end of production and consumer availability. AI Pin’s services, features, and data access will be available till February 28, 2025, 12 pm PST. Grok 3 launched; Musk claims it is the Smartest AI Grok 3, a chatbot built in less than a year, was launched this week in a live demo by the xAI team. The live demonstration showcased Grok 3 handling tasks such as creating a launch plan from Earth to Mars and back and an “insanely great game”, a hybrid between Tetris and Bejeweled. The team claimed that Grok 3’s SOTA model is better than DeepSeek, Claude, and Gemini and is comparable to OpenAI’s model. Check out the recorded demo here (at 19:11 seconds). Project Waterworth, a subsea cable connectivity project by Meta Meta has announced a multi-billion-dollar, multi-year project to open three oceanic corridors connecting five major continents. This will be the longest subsea cable project, spanning 50000 kilometers and linking the U.S., Brazil, South Africa, India, and other key regions. Apart from economic collaboration and digital inclusion, this project aims to drive AI innovation across the world with high-speed connectivity. 💻 Awesome AI: Tools for Work Moonshot AI introduces MoBA that combines Mixture of Experts with sparse attention Following the release of Kimi, Moonshot AI introduced the Mixture of Block Attention (MoBA) model, designed to tackle long conversations and large text. After dividing the text into blocks, MoBA uses a gating mechanism that switches between full and sparse attention, focusing on the most informative blocks, thus reducing computation time. MoBA has been able to maintain competitive performance with 1-million-token context length. Perplexity open-sources DeepSeek R1776 to mitigate bias and censorship To tackle DeepSeek’s avoidance of censored topics in China, Perplexity compiled a dataset of 40k multilingual prompts covering 300 censored topics. R1 was then post-trained on this censorship dataset using an adapted NeMo 2.0 Nvidia framework. The model weights can be downloaded from Hugging Face. Mistral Saba, a custom-trained model for Middle East and South Asian regional languages Mistral has introduced Saba, which has been trained on datasets curated from South Asia and Middle East, to capture cultural and linguistic nuances whilst providing accurate and relevant responses to cater to customers in these regions. Meta Segment Anything Model (SAM) 2.1 is now available in Amazon SageMaker Jumpstart The SOTA vision segmentation model, SAM 2.1, is now publicly available through Amazon SageMaker Jumpstart. SAM 2.1 enables zero-shot object segmentation, object detection using prompts, long-context processing, and context segmentation scenarios. 🛠️ Hackhub Hugging Face introduces agent ratings To evaluate the performance of AI agents in real-world business scenarios, Hugging Face has introduced the AI Agent Leaderboard. The leaderboard currently ranks 17 LLMs, evaluated using the Tool Selection Quality (TSQ) metric across 14 multi-domain datasets. This benchmark assesses LLMs on their ability to select appropriate tools for a given query. This includes parameter handling, multi-step decision making, error handling, context management, and reasoning. At present, gemini-2.0-flash-001 is topping the charts with the highest TSQ of 0.938. LlamaIndex introduces LLM Consortium LlamaIndex has introduced a vision for the AI boardroom of the future by creating an LLM consortium, where multiple LLMs answer the same question, and their responses are synthesized by an arbiter to produce a final result. The arbiter iterates and asks the LLMs to try again if it finds their responses subpar. You can check out the notebook here. Meta achieves breakthroughs in decoding language from brain Meta AI can now decode up to 80% of the characters in a sentence using non-invasive brain recordings. Brain2Qwerty, a deep-learning architecture trained on EEG and MEG data, can decode briefly memorized sentences that participants typed on a QWERTY keyboard. In another related experiment, MEG and EEG data was analyzed to capture the neural dynamics of language production in the human brain. ⚙️Techhub Engine AI’s PM01 Robot deployed for public service in Shenzhen 70 Engine AI’s open-source robots are now serving as community workers and patrolling the streets of Shenzhen, in South China. Powered by DeepSeek, the PM01 robot has achieved human-like mobility and is now making grassroot governance more efficient. YouTube integrates Veo2 to Shorts YouTube Shorts is now integrating Google DeepMind’s popular video generation model. Users in the US, Canada, Australia, and New Zealand can now use text prompts in Shorts to generate standalone video footage. Goku AI ByteDance has recently released GokuAI, a generative flow-based image and video generation model trained on millions of image-text and video-text pairs. Built on a transformed based architecture with 1, 2, and 8 billion parameters, Goku uses diffusion techniques, Rectified Flow, and Variational Autoencoder to create high quality visuals that enable business and content creators to amplify their creative applications. 🧠Masterclass DeepSeek researchers introduce CodeI/O, a new technique to improve LLM reasoning DeepSeek researchers recently shared an approach that uses the structured nature of code to learn symbolic, logical, mathematical, and commonsense reasoning patterns. By collecting Python code from sources like CodeMix and PyEdu-R, the code files are unified using DeepSeek-V2.5. The dataset includes 3.5 million input-output pairs generated from transformed code functions, along with natural language Chain-of-Thought (CoT) explanations. During training, DeepSeek is prompted to generate an output (response), with incorrect responses and feedback fed back into the LLM. Instruction tuning is then applied in the second stage. This multi-turn revision enhances accuracy and shows improvements over baseline models. Less is More for Reasoning (LIMO) improves LLM performance with only 1% training data The LIMO approach challenges the notion that LLMs require extensive data, achieving competitive results with just 817 samples and cognitive templates. LIMO employs a rigorous selection process that includes structural organization, effective cognitive explanations, and verification to curate high-quality math problems from NuminaMath-CoT, AIME, and MATH datasets. Using the Qwen2.5-32B-Instruct model with a 16,384-token sequence length, LIMO applies SFT for training and utilizes step-by-step prompting to achieve generalization capabilities. Large Memory Model (LM2) an auxiliary memory-based model for long context reasoning LM2 incorporates a structured memory system that interacts with input embeddings through cross-attention. Built on a decoder-only transformer architecture, the model utilizes memory updates regulated by gating mechanisms, allowing it to selectively retain relevant information. LM2 was tested on the BABILong and MMLU datasets, demonstrating significant improvements in long-context reasoning and general reasoning capabilities. 📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us. If you have any comments or feedback, just reply back to this email. Thanks for reading and have a great day! *{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Learn all that happened at the AI Action SummitAI_Distilled #82: We are back!AI is not the FUTURE, it’s the PRESENT! Here’s How to NOT Get Left Behind!Want to be ahead of the curve?Block 3 hours of your time to learn AI tools & workflows that 99% of people don’t know yet!🗓️ Tomorrow | ⏱️ 10 AM ESTIn this training, you’ll learn how to:✅ Master 30+ AI tools to automate work & increase efficiency✅ Save 1000s of dollars by leveraging AI for business & personal growth✅ Eliminate repetitive tasks & boost creativity effortlessly✅ Use AI to analyze data, make smarter decisions, and scale fasterHurry! Click here to register (FREE for the first 100 people only!)Hi, there!Greetings for 2025! We’ve been off the radar for a while as we worked on re-inventing our content offerings. AI Distilled will now be run by the LLM Expert Insights team, and we promise to make it up to you with exciting offers in the coming weeks.LLM Expert Insights Team,PacktNewsA two-day AI Action Summit was held in Paris, France on February 10-11, 2025. The summit brought together governments, public and private organizations, academia, NGOs, artists, and civil society. Core themes included public interest AI, the future of work, innovation and culture, trust in AI, and global AI governance. Some of the key announcements were: AI Action Summit Declaration 73 participating members, including 27 EU states, governments, research institutes, and government bodies signed the statement on inclusive and sustainable AI for people and the planet. The UK and the US refrained from signing the declaration. EU launches InvestAI initiative to mobilise €200 billion of investment in artificial intelligence   The InvestAI initiative was announced at the Paris summit with a pledge of EUR 150 billion from the private sector and EUR 50 billion from the public sector. This initiative will support the computing power for the world’s fastest public supercomputers. Ursula von der Leyen, the EU Commission President, vowed in her speech to cut red-tape in AI while ensuring safe AI, encouraging the collaboration of global talent with AI Gigafactories. Launch of public interest initiatives   Current AI, an international partnership of governments, philanthropists, and industry, was officially launched at the AI Action Summit with $400 million in funding, shared Martin Tisné, CEO of AI Collaborative, in his LinkedIn post. Robust Open Online Safety Tools (ROOST)  a non-profit organization incubated at The Institute of Global Politics at Columbia University was also launched at the summit. ROOST has some of the biggest names in the industry as founding partners, including Google, Discord, OpenAI, Roblox and GitHub, Hugging Face, Microsoft, Wikimedia, among others. ROOST aims to provide open-source building blocks and safety resources to global users and communities. OpenAI Roadmap announcedOpen AI will now focus on simplifying product offerings and unify o-series and GPT series models. There will be no o3 release, but GPT-5 will be rolled out with a higher-level intelligence setting for Pro and Plus subscribers and standard intelligence for free tier users.Groq secures $1.5bn from Saudi Arabia to expand AI inference infrastructure in the region Groq CEO Jonathan Ross announced in a LinkedIn post a $1.5 billion agreement to expand Groq’s LPU-based AI infrastructure. This investment will support Groq’s existing data centre in Saudi Arabia and fuel the development of the Arabic Large Language Model (ALLaM).  Elon Musk-Led Group Makes $97.4 Billion Bid for Control of OpenAI, SamA not interested  A group of investors led by Elon Musk has offered to buy control of OpenAI for $97.4 billion. This bid introduces a new twist in OpenAI’s future as the company moves towards restructuring in order to transition to a for-profit entity. The bid backed by xAI, Baron Capital Group, Emanuel Capital Management, 8VC, Valor, Atreides, and Vy Capital is Musk’s latest attempt to make OpenAI open-source and safety-focused, as confirmed by Musk’s attorney, Marc Toberoff. Sam Altman (SamA) took to X to express disinterest in the offer and instead made a counteroffer. 💻 Awesome AI: Tools for WorkMeet New Perplexity Sonar Perplexity has released an optimized version of Sonar to improve decoding throughput which now reaches 1,200 tokens per second. Graded on a scale of 1 to 100, Perplexity’s experiments report that Sonar now scores 85.1 on factuality and 85.9 on readability, surpassing other frontier models. The latest version of Sonar is now available in default search mode for Perplexity Pro users. Cursor’s AI Agent Gets New Capabilities Cursor has added new features to its agent that allow it to accomplish end-to-end development tasks while collaborating with programmers. Some of these features include understanding codebase context, automatically writing and running terminal commands with a programmer’s permission and detecting and fixing lint errors. GitHub Copilot: The agent awakens - The GitHub Blog GitHub unveiled Project Padawan to introduce Copilot’s autonomous agent. In agent mode, Copilot utilizes a SWE agent that can suggest terminal commands, recognize and fix errors, walk through its code, analyse its output and result, debug, diagnose, and fix errors. Apart from this, GitHub also announced the GA of Copilot Edits in VS Code to help developers make inline changes to multiple files in their workspace using natural language. HackhubHugging Face announces AI Energy Score Ratings To drive the adoption of energy-efficient AI, Hugging Face launched the AI Energy Score project. This project offers standardized benchmark for the energy consumption of various AI models. Developers can submit their models to be assessed by a uniform framework and obtain a star rating for their models. There is also a leaderboard that presently ranks 166 models. Go check it out. Open R1 project introduces OpenR1-Math-220k After launching the OpenR1 project to reproduce DeepSeek-R1’s data and training pipeline, the community, in collaboration with Project Numina, announced the construction of OpenR1-Math-220K generated by prompting DeepSeek-R1. Anthropic Economic Index Anthropic analyzed Claude.ai’s anonymized conversations to study how AI is used in real-world tasks and its impact on the labor markets. The study found that 37.2% of conversations were centered around computer and mathematical domains. Computer programmers and copywriters with mid-to-high-median salaries were the highest AI users. The dataset and report have been open sourced. LumaLabsAI drops image to video model In an X post, LumaAI announced the release of image-to-video generation using the Ray2 model. Users subscribed to LITE or PLUS plans can drop any image into the Dream-Machine and create realistic videos. ByteDance introduces OmniHuman-1 ByteDance has released an AI framework that can generate human videos from a single image and motion signal. This diffusion-transformer-based animation framework uses multiple modalities (audio, video, and a combination of signals) to achieve realistic human video generation. TechwaveOpen AI introduces the Intelligence Age with its SuperBowl debut ad To reach the masses, Open AI positioned ChatGPT as the precursor to the Intelligence Age in its first-ever television ad. The ad showcased AI as a tool and brainstorming partner to “assist, aid, and enhance” human-led product vision. Sam Altman’s views on the economics of AI SamA noted in his blog that investing money and resources in AI will drive gains in intelligence for AI models and that the cost of using AI will continue to drop over time, allowing for its wider adoption. He also announced the rollout of AI agents capable of replacing junior level software engineers, potentially impacting jobs and the economy. MasterclassMeta is working on Pippo, a generative model for turnaround videos of humans using a single image Pippo is a multi-view diffusion transformer model pre-trained on 3 billion uncaptioned human images, using both full-reference and cropped versions. It also uses head orientation, position (2D projected anchor), and target camera viewpoint as input.  The model undergoes mid-training on low-resolution images and post-training on high-resolution studio camera images of humans. While the mid-training phase uses an MLP, a ControlNet-inspired MLP is applied to create a 3D-aware multi-view model. Visit here for a visual demo. Decoding-based Regression - Google DeepMind Researchers at DeepMind investigated the use of LLMs to perform regression task by representing numeric predictions as decoded strings and using auto-regressive prediction. They experimented with both normalized and un-normalized tokenization. The proposed approach performed as well as traditional approaches, can be applied to density estimation tasks, and could capture distributions modelled over Gaussian and Riemann distributions. Tell us more about your content needsWe would love to hear from you! Fill out this form to tell us what you’d like to read in AI Distilled next.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Scheduled tasks in ChatGPTAI_Distilled #81: Introducing Microsoft 365 Copilot ChatWorld’s first 16 Hour LIVE Training to become an AI-Powered human in 2025The world of AI is evolving at lightning speed, and the only way to stay relevant is to MASTER AI before it masters you.Join the World’s first 2-Day Mastermind Challenge to learn the Tools, Tactics, and Strategies to Automate Your Work Like Never Before!Best part? It is usually for $395, but the first 100 of you get in for FREE!Claim your FREE spot now!Welcome to AI_Distilled. Today, we’ll talk about:TechwaveCopilot for all: Introducing Microsoft 365 Copilot ChatScheduled tasks in ChatGPTAndrew Ng announces AI-powered Climate SimulatorGitHub Next | Copilot WorkspaceCodestral 25.01 | Mistral AI | Frontier AI in your handsAwesome AI:GitPodcastScrape anything with AI - FetchFoxAISmartCube - Low Code AI ToolsSTORMWhisk by Google LabsMasterclassTitans: Learning to Memorize at Test TimeAgentsHuatuoGPT-o1, Towards Medical Complex Reasoning with LLMsAutoGen v0.4: Reimagining the foundation of agentic AI for scale, extensibility, and robustnessAgent Laboratory: Using LLM Agents as Research AssistantsHackhubfacebookresearch/coconut: Training Large Language Model to Reason in a Continuous Latent SpaceEfficient-Large-Model/Sanavikhyatk/moondream2hexgrad/Kokoro-82MSky-T1: Train your own O1 preview model within $450Cheers,Shreyans SinghEditor-in-Chief, PacktCloud Conversations: A Fireside Chat with Forrest Brazeal and RubrikJoin us on Jan. 28th @ 10 AM PST for a captivating fireside chat where storytelling meets cloud innovation. Forrest Brazeal—acclaimed cloud architect, author, and the creative mind behind cloud computing's most beloved cartoons—teams up with Rubrik’s Chief Business Officer, Mike Tornincasa to explore the evolving challenges of data protection in a multi-cloud world.Save Your Spot⚡ TechWave: AI/GPT News & AnalysisCopilot for all: Introducing Microsoft 365 Copilot ChatMicrosoft has launched Microsoft 365 Copilot Chat, a new AI-powered tool for businesses, combining GPT-4o chat capabilities with agents to automate tasks and enhance productivity. Available in free and pay-as-you-go versions, it allows users to perform tasks like summarizing documents, analyzing data, and generating content while enabling businesses to create custom agents for workflows like customer service and field operations.Scheduled tasks in ChatGPTOpenAI has introduced Scheduled Tasks in ChatGPT, now available in beta for Plus, Pro, and Team users on Web, iOS, Android, and macOS (Windows support coming later). This feature lets users automate tasks by scheduling prompts for specific times or intervals. Tasks run independently of user activity, with notifications sent upon completion. Examples include daily reminders, news briefings, or language practice. Users can manage, edit, or delete tasks through a dedicated "Tasks" menu and customize notification preferences. Limited to 10 active tasks, this beta feature supports GPT-4o capabilities while expanding automation and proactive engagement in ChatGPT workflows.Andrew Ng announces AI-powered Climate SimulatorAndrew Ng recently announced the release of an AI-powered Climate Simulator to explore how geoengineering, specifically Stratospheric Aerosol Injection (SAI), could help mitigate global warming. SAI involves injecting aerosols into the stratosphere to reflect a small portion of sunlight, potentially cooling the planet and opening pathways to limit global warming to 1.5°C. The simulator allows users, including policymakers and the public, to experiment with SAI deployment scenarios and compare their effects against continued warming.GitHub Next | Copilot WorkspaceCopilot Workspace is a developer environment powered by AI, designed to simplify everyday coding tasks. It allows users to describe their goals in natural language, with AI agents proposing and implementing plans, troubleshooting errors, and brainstorming ideas. Features like an integrated terminal, repair suggestions, and easy collaboration make development seamless, while secure versioning and one-click PR creation streamline workflows.Codestral 25.01 | Mistral AI | Frontier AI in your handsCodestral 25.01 is a cutting-edge coding model from Mistral AI, designed to make software development faster and more efficient. Optimized for tasks like code completion, correction, and test generation, it supports over 80 programming languages and excels in fill-in-the-middle (FIM) scenarios. The latest update offers twice the speed of its predecessor, a more efficient architecture, and better tokenizer performance, making it a leader among lightweight coding models.💻 Awesome AI: Tools for WorkGitPodcastGitPodcast is a tool that transforms GitHub repositories into quick, engaging podcasts, making it easier to understand projects on the go. Simply replace "hub" with "podcast" in a GitHub URL to generate a podcast summarizing the repository. It offers short (~5-minute) podcasts for quick insights and longer (~10-minute) versions with a sign-in. This is especially useful for developers and teams who want a convenient way to grasp project details without diving into the code directly.Scrape anything with AI - FetchFoxFetchFox is an AI-powered web scraping tool that lets users extract data from any website by simply describing what they want in plain English. Available as a Chrome extension or npm library, it enables tasks like collecting leads, market research, or analyzing directories.AISmartCube - Low Code AI ToolsAISmartCube is a no-code platform that allows you to build and deploy AI tools easily using drag-and-drop functionality, much like assembling Lego blocks. It offers a wide range of features, including access to large language models like ChatGPT and Claude, integration with plugins for tasks like data scraping, SEO, and image or voice processing, and a real-time shared knowledge base to keep your tools updated. You can automate tasks with ready-to-use templates for social media, copywriting, and e-commerce, or customize AI assistants to handle specific workflows.STORMThe STORM website, developed by Stanford's OVAL lab, is a research preview tool that generates Wikipedia-like reports using AI. Users must agree to terms stating that STORM has limited safety measures, may generate offensive or incorrect content, and should not be used for illegal, harmful, or inappropriate purposes.Whisk by Google LabsWhisk is a new experimental tool from Google Labs that allows users to create and remix images by inputting other images instead of using lengthy text prompts. You can provide a subject image, a scene image, and a style image, and Whisk will combine them into unique creations, such as digital art or merchandise designs. The AI behind Whisk uses the Gemini and Imagen models to process the images and generate new combinations, but it is designed for creative exploration rather than precise edits. The tool is meant to quickly experiment with visual ideas, and users can tweak the results if needed.🔛 MasterclassTitans: Learning to Memorize at Test TimeThe paper introduces "Titans," a new family of neural architectures designed to improve memory handling in machine learning models, addressing challenges of scalability and long-term dependency modeling. Traditional Transformers excel at capturing short-term dependencies but face efficiency issues due to quadratic memory complexity. Titans incorporate a novel neural long-term memory module, inspired by human memory, to memorize past data effectively and complement the short-term memory of attention mechanisms. This architecture integrates three key components: short-term memory for immediate context, long-term memory for persistent historical information, and persistent memory for task-specific knowledge.AgentsIntelligent AI agents are systems designed to perceive and act upon their environment to accomplish tasks, from creating websites to analyzing data. These agents, powered by foundation models, gain enhanced capabilities through tools like knowledge retrievers, web browsers, and code interpreters, allowing them to adapt and perform complex tasks in varied environments. While tools significantly boost their performance, agents face challenges like compounding errors over multiple steps and higher risks due to their ability to perform impactful tasks. Effective agents rely on strong planning capabilities, careful tool selection, and robust security measures to minimize failure modes and ensure reliable, beneficial operation.HuatuoGPT-o1, Towards Medical Complex Reasoning with LLMsHuatuoGPT-o1 is a medical large language model (LLM) designed to excel in complex medical reasoning by leveraging a novel two-stage training process. The approach starts by using a verifier to guide the model in constructing and refining reasoning trajectories for verifiable medical problems, which are derived from challenging medical exam questions. These refined trajectories are used to fine-tune the model. In the second stage, reinforcement learning (RL) with verifier-based feedback further enhances reasoning abilities. This method enables HuatuoGPT-o1 to iteratively analyze and correct its reasoning, achieving superior performance on medical benchmarks compared to general and medical-specific models, all while using only 40,000 training problems.AutoGen v0.4: Reimagining the foundation of agentic AI for scale, extensibility, and robustnessAutoGen v0.4 is a major update to Microsoft's agentic AI framework, enhancing scalability, extensibility, and robustness for multi-agent systems. It introduces an asynchronous, event-driven architecture with modular components, enabling seamless communication, debugging, and observability. The framework supports cross-language compatibility (Python and .NET), robust type enforcement, and distributed agent networks. Key tools include AutoGen Bench for benchmarking and AutoGen Studio, a low-code interface for rapid prototyping with real-time updates, interactive feedback, and visual message flow mapping. Additionally, a new multi-agent application, Magentic-One, tackles complex web and file-based tasks.Agent Laboratory: Using LLM Agents as Research AssistantsAgent Laboratory is an open-source framework that uses large language models (LLMs) to assist researchers in executing machine learning projects efficiently and cost-effectively. It automates key research stages—literature review, experimentation, and report writing—producing comprehensive outputs like research reports and code repositories. Users can provide feedback at each stage, significantly improving output quality. The framework supports various compute levels, making it accessible to different users, and offers a "co-pilot" mode for collaborative research.🚀Hackhubfacebookresearch/coconut: Training Large Language Model to Reason in a Continuous Latent SpaceCoconut is an open-source framework developed by Facebook Research for training large language models (LLMs) to reason in a continuous latent space. It supports end-to-end workflows for research, from preprocessing datasets to training and evaluating models. The framework includes configurations for various reasoning models, like CoT (Chain-of-Thought) and Coconut, with flexible settings for training stages, batch sizes, and checkpoints. Users can customize runs using YAML files and log experiments with wandb. Coconut is designed to reproduce state-of-the-art results on reasoning tasks like GSM8K and ProntoQA, enabling scalable and efficient experimentation with detailed documentation for setup and usage.Efficient-Large-Model/SanaSana is a cutting-edge text-to-image framework developed by NVIDIA that generates high-resolution images up to 4096 × 4096 pixels with remarkable speed and text-image alignment. Based on a Linear Diffusion Transformer architecture with 1648M parameters, it leverages pretrained encoders and advanced diffusion techniques for efficient synthesis. Designed for research and artistic applications, Sana supports creative workflows, educational tools, and the exploration of generative models. While capable of producing stunning visuals, it has limitations in photorealism and handling complex text or detailed features.vikhyatk/moondream2Moondream2 is a compact vision-language model optimized for efficient operation on edge devices, enabling tasks like image captioning, visual querying, object detection, and more. With 1.93 billion parameters and FP16 tensors, it offers advanced features such as streaming caption generation and fine-grained visual understanding. Users can easily integrate it via Hugging Face's Transformers library, with options for GPU acceleration.hexgrad/Kokoro-82MKokoro-82M is a lightweight text-to-speech (TTS) model designed for efficient and high-quality audio generation, featuring just 82 million parameters. Despite its compact size, it has achieved top rankings in the TTS Spaces Arena for single-voice settings, outperforming much larger models in Elo ratings. Kokoro supports American and British English, utilizes an Apache 2.0 license, and offers voice customization through multiple pre-trained voicepacks. Trained on less than 100 hours of permissive audio, Kokoro is optimized for edge devices and is easy to use via ONNX or Python-based workflows. Its design is based on StyleTTS 2 and ISTFTNet architectures, prioritizing accessibility and efficiency.Sky-T1: Train your own O1 preview model within $450NovaSky, a team from UC Berkeley's Sky Computing Lab, developed Sky-T1-32B-Preview, an open-source reasoning model trained for under $450. This model rivals proprietary reasoning models like o1-preview in tasks like math and coding, while being fully transparent with its data, code, and weights. By refining training methods, balancing diverse datasets, and leveraging efficient infrastructure, NovaSky enables the academic and open-source community to replicate and improve upon their results.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Devin is now generally availableAI_Distilled #80: Google introduces Gemini 2.0: A new AI model for the agentic eraZapier connects the apps you use every day, so you can focus on what matters most - Free to start.Learn MoreWelcome to AI_Distilled. Today, we’ll talk about:TechwaveDevin is now generally availableGoogle introduces Gemini 2.0: A new AI model for the agentic eraMeta Llama-3.3 70B-InstructGemini Flash - Google DeepMindI can now run a GPT-4 class model on my laptopAwesome AI:Retro Diffusion: The Future of Pixel Art is nowMagic Clips: Create Viral Clips From Long Videos, Instantlysoundfont-generator - a Hugging Face Space by erl-jPickle -Lifelike AI clones lip-syncing to your voice in real-timeShortcut by Poised12 days of OpenAI:Day 5: Apple launches its ChatGPT integration with SiriDay 4: OpenAI Canvas Kills Google Docs, Challenges VS Code & CursorDay 3: OpenAI has finally released SoraDay 2: Reinforcement Fine-Tuning Research ProgramDay 1: Introducing ChatGPT ProSecret Knowledge:Hugging Face's Text Generation Inference v3 overviewMeet Willow, our state-of-the-art quantum chipGrok Image Generation ReleaseThis is our final edition of AI_Distilled for 2024, but don’t worry—we’ll be back with more insights and updates in January 2025. In the meantime, we’ve got a little holiday treat for you!Packt has some exciting offers lined up to help you boost your tech skills and get ready for an amazing new year! It’s the perfect opportunity to relax, learn something new, and stay ahead in your field. Keep an eye out for these special holiday deals!From all of us at the Packt Newsletters team, we wish you a joyful holiday season and a fantastic start to 2025. See you next year!Cheers,Shreyans SinghEditor-in-Chief, PacktStop worrying about your to-do list.Zapier connects the apps you use every day, so you can focus on what matters most.Start working more efficiently - Create your free account today.Get started for free⚡ TechWave: AI/GPT News & AnalysisDevin is now generally availableDevin, a powerful AI tool for engineering teams, is now generally available starting at $500 per month. With no seat limits, integrations for Slack, IDEs, and APIs, and direct support from Cognition's engineering team, Devin is designed to tackle small frontend bugs, create first-draft PRs, and perform targeted code refactors. Teams can collaborate with Devin via Slack for task management, use its IDE extension for code reviews, and guide it with feedback to refine its output.Google introduces Gemini 2.0: A new AI model for the agentic eraGoogle unveiled Gemini 2.0, its next-generation AI model, designed for "agentic" capabilities, enabling AI to act proactively on behalf of users. The multimodal model can process and generate text, images, audio, and video while using tools like Google Search and code execution. Its experimental version, Gemini 2.0 Flash, is available to developers with enhanced performance and lower latency.Meta Llama-3.3 70B-InstructLlama 3.3 is a powerful multilingual AI model developed by Meta, designed for generating text and assisting in conversations across multiple languages. With 70 billion parameters, it uses advanced transformer architecture and aligns with human preferences through fine-tuning methods like RLHF. The model supports multilingual text input and output, offering robust performance in tasks like coding, reasoning, and multilingual understanding. It incorporates a long context window, tool use capabilities, and optimized inference using Grouped-Query Attention.Gemini Flash - Google DeepMindGemini 2.0, developed by Google DeepMind, is a cutting-edge AI model designed for a new era of "agentic" experiences, where AI systems can perform tasks using memory, reasoning, and planning under human supervision. This model features enhanced capabilities like native tool usage, real-time multimodal understanding (text, images, video, and audio), image generation, and text-to-speech. It excels in low-latency scenarios, enabling applications like coding assistance, game navigation, and interactive learning experiencesI can now run a GPT-4 class model on my laptopMeta’s Llama 3.3 70B is a groundbreaking language model that matches GPT-4’s capabilities and can run on consumer-grade laptops like a 64GB MacBook Pro M2. This remarkable feat showcases the rapid advances in AI model efficiency over the past two years, making high-quality AI tools more accessible than ever. By using tools like Ollama, users can now easily download and run these models locally, enabling powerful applications like text generation and coding assistance. The model has also performed competitively on benchmarks, cementing its position among leading LLMs. This progress highlights the potential for affordable, locally hosted AI, expanding its utility for developers and enthusiasts alike.💻 Awesome AI: Tools for WorkRetro Diffusion: The Future of Pixel Art is nowRetro Diffusion is a cutting-edge platform designed by artists to simplify and enhance the process of creating pixel art. It offers specialized tools that eliminate common frustrations, enabling creators to focus on their artistry rather than technical hurdles. With Retro Diffusion, artists can quickly achieve professional-level pixel art, transforming their creative visions with ease and efficiency.Magic Clips: Create Viral Clips From Long Videos, InstantlyMagic Clips is an AI-powered platform that transforms long videos into engaging, viral short clips instantly without the need for manual editing. Simply upload a video or paste a link, and the AI selects the most captivating moments, adds captions, and arranges them into shareable content. With features like unlimited uploads, transcript navigation, and performance optimization, Magic Clips helps users create and repurpose content efficiently.soundfont-generator - a Hugging Face Space by erl-jErl-j's Soundfont Generator is an AI tool that creates custom soundfonts based on text descriptions. Users simply input a prompt describing the desired audio (e.g., "hard bass" or "sparkly bells"), adjust the generation settings for quality or speed, and generate the soundfont. The tool allows users to preview the instrument using a virtual keyboard and export it as a downloadable SFZ soundfont package, compatible with SFZ samplers. Built on advanced audio models, it uses latent flow matching for faster and efficient generation, making it a powerful resource for musicians and audio designers.Pickle - Lifelike AI clones lip-syncing to your voice in real-timePickle lets you use a personalized AI clone to represent you in video calls, providing flexibility and freedom. Whether you're not camera-ready, multitasking, or taking a break, your clone seamlessly participates in meetings across any video platform. With customizable outfits and backgrounds, you can tailor your clone to suit your needs.Shortcut by PoisedShortcut is an AI-powered tool that transforms the way you work by enabling natural voice-based interaction instead of typing. It lets you ask questions, organize ideas, and create polished drafts of messages, emails, and documents instantly, maintaining your productivity flow. With Shortcut, your spoken words are quickly refined into well-crafted text in your chosen tone—friendly, professional, or direct—eliminating the hassle of editing.🔛 12 days of OpenAIDay 5: Apple launches its ChatGPT integration with SiriApple has launched ChatGPT integration with Siri as part of its new iOS 18.2 update, enabling Siri to handle complex questions by seamlessly accessing OpenAI’s GPT-4o model with user permission. This marks a significant step in Apple's AI initiative, dubbed Apple Intelligence, which aims to enhance user experience with advanced tools like text rewriting, glowing Siri notifications, and app action capabilities coming next year. The integration prioritizes privacy, ensuring OpenAI doesn’t store user queries, and positions Apple as a leader in consumer AI while offering OpenAI exposure to millions of iPhone users.Day 4: OpenAI Canvas Kills Google Docs, Challenges VS Code & CursorOpenAI has introduced Canvas, a new feature within ChatGPT that provides a split-screen interface for drafting, editing, and coding, aiming to compete with tools like Google Docs, VS Code, and Cursor. Users can write or code on one side while receiving real-time suggestions and feedback from ChatGPT on the other. This feature supports Python code execution, debugging, and syntax highlighting, making it a robust tool for developers and writers alike. Beyond basic editing, users can format text, address AI-generated comments, and generate visual outputs using Python.Day 3: OpenAI has finally released SoraOpenAI has launched Sora, a groundbreaking text-to-video AI tool, offering users the ability to create 1080p videos up to 20 seconds long with the $200/month ChatGPT Pro subscription, or shorter 720p videos with ChatGPT Plus. Users can generate videos from text, animate images, remix existing videos, and even blend scenes with AI. Sora includes features like a storyboard tool for precise frame-by-frame input and a community feed showcasing creations. All videos come with watermarks and metadata to ensure transparency and prevent misuse.Day 2: Reinforcement Fine-Tuning Research ProgramOpenAI has launched the Reinforcement Fine-Tuning Research Program to enable developers and machine learning engineers to fine-tune AI models for domain-specific tasks. This technique involves training models using curated high-quality tasks and grading their responses against reference answers, improving reasoning and accuracy in specific fields like law, healthcare, and finance. Participants in the program gain alpha access to the Reinforcement Fine-Tuning API to test its potential on their use cases and provide feedback ahead of its public release in 2025.Day 1: Introducing ChatGPT ProOpenAI has introduced ChatGPT Pro, a premium subscription plan costing $200 per month, offering enhanced access to its most advanced AI models and tools. This includes the powerful o1 Pro mode, which uses increased computational resources to provide more accurate and comprehensive answers, especially for complex tasks in data science, programming, and advanced research. External evaluations highlight its superior performance across challenging benchmarks like competitive math, coding, and science problems.🚀 Secret KnowledgeHugging Face's Text Generation Inference v3 overviewHugging Face's Text Generation Inference (TGI) v3 delivers significant performance enhancements for handling large language models (LLMs). It processes three times more tokens and is 13 times faster than its competitor vLLM for long prompts, thanks to optimized memory usage, efficient prefix caching, and streamlined configurations that require no manual setup. TGI also improves hardware utilization, making it adaptable for both small-scale and high-performance deployments. Benchmarks confirm these gains across various scenarios, showcasing faster responses for long conversations and complex prompts.Meet Willow, our state-of-the-art quantum chipGoogle's latest quantum chip, Willow, represents a significant leap forward in quantum computing, addressing long-standing challenges in error correction and performance. Willow demonstrates the ability to reduce errors exponentially as more qubits are added, solving a decades-old problem in quantum error correction. It also performed a benchmark computation in under five minutes, a task that would take the fastest classical supercomputers 10 septillion years, highlighting its unmatched processing power. With 105 qubits and breakthroughs in chip design, Willow is a major milestone toward building large-scale, practical quantum computers capable of tackling real-world problems and advancing scientific discovery.Grok Image Generation ReleaseGrok's new image generation model, Aurora, brings cutting-edge capabilities to the 𝕏 platform, offering photorealistic rendering and precise adherence to text prompts. Trained on billions of text and image examples, Aurora supports multimodal input, enabling users to generate original images, edit existing ones, and create artistic or realistic visuals with exceptional detail. Its versatility spans entity creation, artistic designs, and realistic human portraits.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Google announces Veo and Imagen 3: new video and image generation modelsAI_Distilled #79: Sam Altman announces "12 days of OpenAI"Learn Million Dollar AI Strategies & Tools in this 3 hour AI Training for Free.If you are not an AI-powered professional today, you will either:-Get replaced by a person who uses AI-Face a slow career growth & lower salary-Keep spending 10s of hours on tasks that can be done in 10 minutes.Best thing? We’re running the Black Friday Sale so you can get it for absolutely free (for the first 100 readers).Save your seat now (Offer valid for 24 hours only)Welcome to AI_Distilled. Today, we’ll talk about:TechwaveSam Altman announces "12 days of OpenAI"Google announces Veo and Imagen 3: new video and image generation modelsDeepMind Genie 2: generate interactive worlds that look like video gamesIntel data scientist's survival guide to GenAINvidia launches Ingest: Multimodal PDF Data ExtractionAwesome AI:Polymet - Idea to prototype within secondsClipAnything - Choppityfal.aiEarkick - Your Personal AI ChatbotOuterbase | The interface for your databaseMasterclass:Voice Trigger System for SiriAlign Meta Llama 3 to human preferences with DPOAn Intuitive Intro to RLEnhancing LLMs with Structured Outputs and Function CallingSafely repairing broken builds with MLHackHub:Agents for software developmentOpen-source LLM app development platformbuild, manage & run useful autonomous agentsUnderstand Human Behavior to Align True NeedsGenerative models for conditional audio generationCheers!Shreyans SinghEditor-in-Chief, Packt⚡ TechWave: AI/GPT News & AnalysisSam Altman announces "12 days of OpenAI"OpenAI is celebrating with a special event called "12 Days of OpenAI," where, for twelve days, the company will reveal new models, features, and updates via livestreams. Anticipated reveals include full release of its o1 reasoning model, updates on its voice modes, including a festive Santa voice, a new AI agent called Operator, a web browser, a desktop app update, and advancements in AI-generated music and vision fine-tuning. Notably, OpenAI may also introduce new AI chips and even GPT-5, which promises improved reasoning and customization.Google announces Veo and Imagen 3: new video and image generation modelsGoogle Cloud has introduced two advanced generative AI models, Veo and Imagen 3, on its Vertex AI platform. Veo allows businesses to generate high-quality videos from simple text or image prompts, transforming creative assets into dynamic visuals quickly and affordably. Imagen 3, launching next week, creates highly realistic images from text prompts, offering more detail and fewer visual artifacts than previous models. Both models are built with safety features, such as digital watermarking and safety filters, to ensure responsible use.DeepMind Genie 2: generate interactive worlds that look like video gamesDeepMind has introduced Genie 2, an advanced AI model capable of generating interactive 3D worlds that resemble video games. Unlike previous models, Genie 2 can create dynamic environments from just a single image and a text description, allowing users to interact with the scene, like jumping or swimming. The model simulates object interactions, physics, and animations, and can remember parts of the world even when they’re not visible, offering a more consistent and realistic experience. While not designed for full gaming experiences, Genie 2 is a tool for research, creative prototyping, and evaluating AI agents.Intel data scientist's survival guide to GenAIWhile GenAI tools can produce impressive results, they heavily rely on clean, well-structured data and insightful interpretation—areas where data scientists excel. Your expertise in data analysis, modeling, and statistical methods ensures that these models can make accurate, actionable predictions. GenAI platforms need data scientists to optimize and evaluate models, enhance their performance, and ensure their deployment is successful. Tools like Modin, Intel-optimized frameworks, and MLflow help streamline the process, making data preparation, model training, and deployment more efficient, particularly when working on Intel hardware.Nvidia launches Ingest: Multimodal PDF Data ExtractionNVIDIA-Ingest is a powerful microservice for extracting and processing content from documents like PDFs, Word, and PowerPoint files. It can analyze and separate text, images, tables, and charts, delivering them in a structured JSON format. Using NVIDIA's advanced tools, including OCR and AI-driven parsing, it enables efficient data processing for downstream applications like generative AI or embedding storage in vector databases like Milvus. It supports flexible workflows and can handle tasks like splitting documents, generating embeddings, and transforming data💻 Awesome AI: Tools for WorkPolymet - Idea to prototype within secondsPolymet is an AI-powered tool that helps users quickly turn ideas into prototypes by generating designs and production-ready code in seconds. Users can describe what they need, iterate on the design with their team, and then export the code and designs, which can easily integrate with tools like Figma and existing codebases.ClipAnything - ChoppityChoppity is an AI-powered video editing tool that allows users to quickly find and clip moments from any video using visual, audio, and sentiment analysis. With its "ClipAnything" feature, users can search for specific parts of a video, such as key events, people, or emotions, without having to manually review hours of footage.fal.aiFal.ai is a generative media platform designed for developers to create and deploy AI-powered applications, particularly focused on text-to-image models. It offers fast, cost-effective inference with models like FLUX.1 and Stable Diffusion, optimized for various creative tasks.Earkick - Your Personal AI ChatbotEarkick is an AI-powered mental health app that helps users track and improve their emotional well-being in real time through a personal chatbot named Panda. Earkick tracks mental readiness, mood, and calmness, while providing daily insights, breathing techniques, and guided self-care sessions.Outerbase | The interface for your databaseOuterbase is an AI-powered platform that simplifies working with databases for engineers, researchers, and analysts. It supports SQL and NoSQL databases, allowing users to manage data securely while using AI tools to write queries, fix mistakes, and generate charts and visualizations instantly. Outerbase's table editor, dashboards, and data catalog help users organize, analyze, and share insights efficiently.🔛 Masterclass: AI/LLM TutorialsVoice Trigger System for SiriApple's voice trigger system for Siri includes a first-stage low-power detector to identify potential triggers, and a second-stage, high-precision model to confirm the trigger. It also incorporates speaker identification to ensure the device responds only to its primary user. This sophisticated setup addresses challenges like background noise and phonetically similar words while maintaining power efficiency and privacy.Align Meta Llama 3 to human preferences with DPODPO involves fine-tuning a large language model (LLM) based on feedback from human annotators who rate or rank the model's responses according to desired values, such as helpfulness and honesty. SageMaker Studio provides the computational environment to fine-tune the model using Jupyter notebooks with powerful GPU instances, while SageMaker Ground Truth simplifies the process of gathering human feedback by managing workflows for data annotation. Together, they allow you to align the Llama 3 model’s responses with specific organizational values efficiently.An Intuitive Intro to RLReinforcement learning (RL) is a type of machine learning where an agent learns by interacting with its environment, making decisions, and receiving feedback in the form of rewards or penalties. The goal is to maximize cumulative rewards over time. The agent starts with little to no knowledge and improves through trial and error, learning from past experiences. In RL, actions taken by the agent change the state of the environment, and based on the rewards received, the agent adjusts its future actions. A key concept in RL is balancing exploration (trying new things) and exploitation (using known strategies for rewards).Enhancing LLMs with Structured Outputs and Function CallingEnhancing LLMs with structured outputs and function calling improves their ability to provide accurate and useful responses. Structured outputs ensure consistency and clarity by organizing information in a logical format, reducing ambiguity. Function calling allows LLMs to perform specific tasks, such as retrieving real-time data or executing external functions, making them more interactive and versatile. Combined with techniques like Retrieval-Augmented Generation (RAG), which integrates relevant external information into the model’s responses, these enhancements lead to more reliable, accurate, and contextually rich conversations with LLMs.Safely repairing broken builds with MLGoogle's engineers have developed a machine learning model called DIDACT to automatically repair broken code builds by analyzing historical data of build errors and their fixes. This model suggests potential fixes to developers directly within their Integrated Development Environment (IDE). In a controlled experiment, the use of these machine learning-suggested fixes improved productivity by reducing active coding and feedback time, and increasing the number of completed code changes.🚀 HackHub: AI ToolsAll-Hands-AI/OpenHandsOpenHands is an AI-powered platform designed to assist with software development, allowing agents to perform tasks similar to human developers. These agents can modify code, run commands, browse the web, call APIs, and even use resources like StackOverflow. OpenHands is easy to set up using Docker and can be run in various modes, including scriptable or interactive CLI.langgenius/difyDify is an open-source platform for developing AI applications, offering an intuitive interface that integrates workflows, agent capabilities, model management, and observability features. Dify's core features include a visual AI workflow builder, integration with numerous LLMs, agent tools, and a retrieval-augmented generation (RAG) pipeline for document handling.TransformerOptimus/SuperAGISuperAGI is an open-source framework designed for developers to create, manage, and run autonomous AI agents. It allows seamless operation of multiple agents simultaneously and provides tools to extend their capabilities. With features like graphical interfaces, performance telemetry, and integration with multiple vector databases, SuperAGI enables AI agents to efficiently handle tasks, learn from experience, and optimize token usage.lllyasviel/Paints-UNDOPaints-Undo is an open-source project that provides AI models designed to simulate the drawing process in digital art. By inputting a completed image, users can generate a sequence of steps showing how that image might have been created, mimicking the "undo" function in digital painting software.Stability-AI/stable-audio-toolsStable-Audio-Tools is an open-source library for working with audio generation models. It provides tools for training and running models that generate audio, including a Gradio interface for testing. Users can install the library via PyPI, and the repository includes scripts for both training models and performing inference.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

AI Code Review for Developers | TragAI_Distilled #78: Customize how Claude responds: Concise, Explanatory, or FormalLearn the Roadmap to making $100k using LinkedIn & AI (for free)🚀In just 90 minutes, you’ll learn how to:👉 Automate lead generation to grow your business effortlessly.👉 Master LinkedIn's $100K strategy to increase revenue while saving time.👉 Use AI to secure high-paying roles, bypassing endless applications.Join Vaibhav Sisinty, a LinkedIn influencer with over 400K followers, who’s transformed the LinkedIn strategies of over 200,000 professionals. Normally valued at $399, this workshop is free for the first 100 readers.Claim Your Free Spot Now (Only 100 seats available!)Welcome to AI_Distilled. Today, we’ll talk about:TechwaveCustomize how Claude responds: Concise, Explanatory, or FormalRunwayML: Introducing FramesAnthropic introduces the Model Context Protocol: SmolVLM - small yet mighty Vision Language ModelCursor announces new code editor UI and agentAwesome AI:Paperguide: AI Research Assistant & Chat with PDFCapGo AI: Spreadsheet That Fills ItselfAI Code Review for Developers | TragConversational AI Survey with Real-time Follow upsSagaLabs: Earn 200x More with In-context AI translation from the worldMasterclass:ControlNets for Stable Diffusion 3.5 Large — Stability AIAutomatically generating cloud configurations: Introducing RAGformationBoost your Continuous Delivery pipeline with Generative AI | Google CloudCreating with Video to Video on Gen-3 Alpha and Turbo – RunwayModel-Based Transfer Learning for Contextual Reinforcement LearningHackHub:Andrew Ng releases an open-source Python framework to swap between LLMs with one line of codeOpenInterpreter/open-interpreter: A natural language interface for computersItzCrazyKns/Perplexica: Perplexica is an AI-powered search engine. It is an Open source alternative to Perplexity AIsouzatharsis/podcastfy: An Open Source Python alternative to NotebookLM's podcast feature: Transforming Multimodal Content into Captivating Multilingual Audio Conversations with GenAIblack-forest-labs/flux: Official inference repo for FLUX.1 modelsCheers!Shreyans SinghEditor-in-Chief, PacktScale your scrapers with Apify’s Black Friday Boost planGet a 30% prepaid usage bonus on Apify this Black Friday. Scrape data for app integrations, performance tracking, competitive research, or custom pipelines. Use pre-built scrapers, build your own from scratch, or use quick-start code templates. The Boost plan ends December 5 - grab it while you can!Claim your bonus now⚡ TechWave: AI/GPT News & AnalysisCustomize how Claude responds: Concise, Explanatory, or FormalAnthropic has introduced a new feature for its Claude AI assistant that allows users to customize its writing style to match their own or adjust it for specific tasks. Users can choose from three preset styles—Formal, Concise, and Explanatory—or create personalized styles by uploading sample text for Claude to mimic. This feature aims to make interactions feel more natural and tailored, whether for technical documents, professional emails, or casual chats.RunwayML: Introducing FramesRunway's new image generation model, Frames, offers advanced stylistic control and visual fidelity, allowing creators to design consistent yet creatively flexible visuals. Integrated into Gen-3 Alpha and the Runway API, Frames helps users craft detailed aesthetic worlds, from cinematic portraits to retro-inspired designs. Frames aims to redefine creative workflows by enabling precise and imaginative visual storytelling.Anthropic introduces the Model Context Protocol: Anthropic has introduced the Model Context Protocol (MCP), an open-source standard aimed at improving how AI assistants access and use data from various sources, like business tools and content repositories. MCP enables two-way connections between AI models and data systems through "MCP servers" and "MCP clients," simplifying integration and reducing the need for custom connectors. promising to create more seamless and scalable AI integrations, MCP faces competition from proprietary alternatives like OpenAI’s "Work with Apps,".SmolVLM - small yet mighty Vision Language ModelSmolVLM is a highly efficient and compact 2-billion-parameter Vision-Language Model (VLM) that delivers state-of-the-art performance for its size and memory usage. Designed for speed, memory efficiency, and ease of customization, SmolVLM is fully open-source under the Apache 2.0 license, with tools, training recipes, and datasets readily available. Its three variants—Base, Synthetic, and Instruct—support fine-tuning and out-of-the-box applications. By optimizing image token encoding and leveraging innovative architecture, SmolVLM runs effectively on smaller devices like laptops, offering fast inference and low GPU memory usage.Cursor announces new code editor UI and agentCursor's 0.43 update transforms the AI-powered code editor into a more efficient and developer-friendly tool. Key features include a unified workspace with the redesigned Composer UI, advanced automation for debugging and package installation via the Composer Agent, and enhanced semantic search for faster, context-aware results. The update also introduces proactive debugging with the experimental BugFinder tool, visual cues for easier file management, and context-aware coding suggestions.💻 Awesome AI: Tools for WorkPaperguide: AI Research Assistant & Chat with PDFCapGo AI: Spreadsheet That Fills ItselfAI Code Review for Developers | TragConversational AI Survey with Real-time Follow upsSagaLabs: Earn 200x More with In-context AI translation from the world🔛 Masterclass: AI/LLM TutorialsControlNets for Stable Diffusion 3.5 Large — Stability AIStable Diffusion 3.5 Large introduces three new ControlNets—Blur, Canny, and Depth—designed to enhance image generation precision. Blur enables high-fidelity upscaling for detailed visuals, Canny uses edge maps for structured illustrations, and Depth leverages depth maps for architectural and 3D applications. These models are free for non-commercial and small-scale commercial use.Automatically generating cloud configurations: Introducing RAGformationRAGformation is an open-source AI tool designed to simplify cloud configuration by automating the selection of services, cost estimation, and architecture design. Using natural language input, it generates tailored cloud setups, including visual flow diagrams, pricing details, and a comprehensive blueprint. Powered by Retrieval-Augmented Generation (RAG) and tools like LlamaIndex and Pinecone, RAGformation dynamically adjusts recommendations based on user preferences and budgets.Boost your Continuous Delivery pipeline with Generative AI | Google CloudGenerative AI, such as Google Cloud's Gemini models, enhances software development by automating repetitive tasks and improving code quality throughout the development lifecycle. Beyond assisting in coding within IDEs, AI can streamline continuous delivery pipelines by automating code reviews, generating release notes, and detecting potential issues early. For example, integrating Gemini into a CI/CD pipeline allows developers to receive AI-driven feedback on pull requests and summaries of code changes, reducing manual effort and boosting productivity. Tools like the "friendly-cicd-helper" demonstrate how AI can complement traditional processes, freeing developers to focus on strategic tasks while maintaining high-quality standards.Creating with Video to Video on Gen-3 Alpha and Turbo – RunwayThe Gen-3 Alpha and Turbo models offer an enhanced "Video to Video" feature, allowing users to transform the style of videos using text prompts. The Turbo model is faster and more cost-effective, supporting resolutions up to 1280x768 and videos of up to 20 seconds. To use this feature, select a model, upload a supported video, and draft a detailed prompt to define the desired style. Additional settings, like structure transformation and aspect ratio, allow for customization. Once configured, the tool generates stylized videos, with results saved in the Generative Video folder for easy access.Model-Based Transfer Learning for Contextual Reinforcement LearningThis paper introduces Model-Based Transfer Learning (MBTL), a framework to improve generalization in contextual reinforcement learning (RL). Traditional RL approaches often fail with minor environmental changes, and existing training methods are either too resource-intensive or prone to negative transfer. MBTL addresses this by modeling generalization performance with Gaussian processes and linear functions to predict and minimize performance gaps when transferring policies to new tasks. By integrating these models with Bayesian optimization, MBTL strategically selects training tasks, achieving up to 50x better sample efficiency in benchmarks like urban traffic. This approach paves the way for more reliable and efficient RL training methods.🚀 HackHub: AI ToolsAndrew Ng releases an open-source Python framework to swap between LLMs with one line of codeOpenInterpreter/open-interpreter: A natural language interface for computersItzCrazyKns/Perplexica: Perplexica is an AI-powered search engine. It is an Open source alternative to Perplexity AIsouzatharsis/podcastfy: An Open Source Python alternative to NotebookLM's podcast feature: Transforming Multimodal Content into Captivating Multilingual Audio Conversations with GenAIblack-forest-labs/flux: Official inference repo for FLUX.1 models📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

What is the Chinchilla Scaling Law?AI_Distilled #77: GenAI for YouTubersWelcome to AI_Distilled. Today, we’ll talk about:Awesome AI:Adobe Firefly Video Model previewReddit ScoutIlluminate by GoogleThunderbit | Personalized Web AI CopilotVerse: Make free digital pagesMasterclass:GenAI for YouTubers- Google DeepMindThe Basics Behind AI Models for Self-Driving CarsWhat is the Chinchilla Scaling Law?Improve RAG performance using Cohere RerankMIT researchers have developed "Co-LLM"HackHub:Upscayl: free and open source AI image upscalerRoop: one-click face swapAnthropic-quickstarts: build deployable applications using the Anthropic APIMulti-GPT: An experimental open-source attempt to make GPT-4 fully autonomousFacebook Audioseal: Localized watermarking for AI-generated speech audiosCheers!Shreyans SinghEditor-in-Chief, Packt💻 Awesome AI: Tools for WorkAdobe Firefly Video Model previewAdobe has introduced its new Firefly Video Model, a generative AI tool designed to enhance video editing within Adobe's software like Premiere Pro. It enables users to generate videos using text prompts, create atmospheric elements like fire or water, fill timeline gaps, and even bring still images to life.Reddit ScoutReddit Scout is a tool that quickly summarizes Reddit comments to help users find the best products to buy, saving time sifting through lengthy threads. It provides a detailed summary of discussions on various topics, such as smart home security systems, and is available as a Chrome extension.Illuminate by GoogleThis platform offers AI-generated audio discussions on various topics, transforming written content into engaging audio summaries. Each entry provides a concise audio summary of key papers and articles, making complex information easily accessible.Thunderbit | Personalized Web AI CopilotThunderbit is an AI-powered tool designed to help business users automate various web tasks. It offers features like AI Web Clipper for extracting essential details from websites, voice note-taking to convert voice into structured notes, and AI-assisted data sync between business tables.Verse: Make free digital pagesVerse is an app that turns your music taste into a visual representation of your personal space, like a digital bedroom inspired by the songs you listen to. It lets you explore and download creative content, from music and art to guides and reviews.🔛 Masterclass: AI/LLM TutorialsEmpowering YouTube creators with generative AI - Google DeepMindGoogle DeepMind is introducing generative AI tools, Veo and Imagen 3, to YouTube creators through a feature called Dream Screen. This will allow users to generate creative video backgrounds for YouTube Shorts by starting with a text prompt and choosing from four AI-generated images. Veo will then turn the selected image into a high-quality 6-second video clip.The Basics Behind AI Models for Self-Driving CarsThis article explains how AI models for self-driving cars work by simulating driving behaviors using sensor data and a neural network. It outlines the basic mechanics: cars are equipped with sensors that detect proximity to objects in all directions, and the model uses this data to predict acceleration, braking, and steering. The neural network is trained on synthetic data that mimics human driving decisions, such as how much to turn or accelerate based on obstacles. A five-layer neural network built with PyTorch is used to train the model, which is evaluated based on its accuracy and crash rates.What is the Chinchilla Scaling Law?The Chinchilla Scaling Law, introduced in 2022, proposes that smaller language models can outperform larger ones if trained on significantly more data. Traditional models like GPT-3 increased in size without proportionally scaling the training data, leading to inefficiencies. The Chinchilla Scaling Law suggests an optimal balance between model size and data, showing that doubling the amount of data for every doubling of model size can maximize performance with the same compute resources.Improve RAG performance using Cohere RerankCohere Rerank helps improve RAG's performance by reordering retrieved documents based on a relevance score using deep learning. This second-stage process refines the results by aligning them more closely with user queries, boosting search accuracy and efficiency. Cohere Rerank can be integrated easily with tools like Amazon SageMaker.MIT researchers have developed "Co-LLM"MIT researchers have developed "Co-LLM," an algorithm that enables large language models (LLMs) to collaborate for more accurate and efficient solutions. It pairs a general-purpose model with a specialized expert model, with a "switch variable" that identifies when the general model needs help. This process allows the general model to handle most of the response, while the expert model steps in only when needed, improving accuracy and efficiency. The approach mimics how humans consult experts for specific tasks.🚀 HackHub: AI Toolsupscayl/upscaylUpscayl is a free, open-source AI-powered image upscaler that lets you enhance and enlarge low-resolution images without losing quality. The tool uses advanced AI algorithms like Real-ESRGAN. You'll need a Vulkan-compatible GPU for best results.s0md3v/roopRoop is an AI-based face-swapping tool that allows you to replace the face in a video with a face of your choice using just a single image—no training or large datasets required. Once set up, you can swap faces in videos by specifying source and target files through command-line options.anthropics/anthropic-quickstartsAnthropic Quickstarts is a set of projects that help developers easily build and deploy applications using the Anthropic API. These quickstarts offer a solid foundation for various applications, starting with a customer support agent powered by Claude, Anthropic's AI.sidhq/Multi-GPTMulti-GPT is an experimental system where multiple specialized GPT models, known as "ExpertGPTs," work together to accomplish tasks. Each expert has its own memory (both short and long-term) and can communicate with other experts to solve complex problems. The system integrates advanced capabilities like internet searches, file storage, and long-term data recall. Users can interact with it by setting tasks, and the experts will collaborate autonomously to complete them, leveraging GPT-4 for text generation and optional tools like Pinecone for memory storage.facebookresearch/audiosealAudioSeal is a speech watermarking method that embeds invisible watermarks into audio, making it possible to detect watermarked segments even after editing. It uses a generator to create watermarks and a detector to find them in real-time with high accuracy, operating up to 100 times faster than existing models.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

An Intuitive Intro to RLAI_Distilled #76: Align Meta Llama 3 to human preferences with DPOWelcome to AI_Distilled. Today, we’ll talk about:Awesome AI:Polymet - Idea to prototype within secondsClipAnything - Choppityfal.aiEarkick - Your Personal AI ChatbotOuterbase | The interface for your databaseMasterclass:Voice Trigger System for SiriAlign Meta Llama 3 to human preferences with DPOAn Intuitive Intro to RLEnhancing LLMs with Structured Outputs and Function CallingSafely repairing broken builds with MLHackHub:Agents for software development Open-source LLM app development platformbuild, manage & run useful autonomous agentsUnderstand Human Behavior to Align True NeedsGenerative models for conditional audio generationCheers!Shreyans SinghEditor-in-Chief, Packt💻 Awesome AI: Tools for WorkPolymet - Idea to prototype within secondsPolymet is an AI-powered tool that helps users quickly turn ideas into prototypes by generating designs and production-ready code in seconds. Users can describe what they need, iterate on the design with their team, and then export the code and designs, which can easily integrate with tools like Figma and existing codebases.ClipAnything - ChoppityChoppity is an AI-powered video editing tool that allows users to quickly find and clip moments from any video using visual, audio, and sentiment analysis. With its "ClipAnything" feature, users can search for specific parts of a video, such as key events, people, or emotions, without having to manually review hours of footage.fal.aiFal.ai is a generative media platform designed for developers to create and deploy AI-powered applications, particularly focused on text-to-image models. It offers fast, cost-effective inference with models like FLUX.1 and Stable Diffusion, optimized for various creative tasks.Earkick - Your Personal AI ChatbotEarkick is an AI-powered mental health app that helps users track and improve their emotional well-being in real time through a personal chatbot named Panda. Earkick tracks mental readiness, mood, and calmness, while providing daily insights, breathing techniques, and guided self-care sessions.Outerbase | The interface for your databaseOuterbase is an AI-powered platform that simplifies working with databases for engineers, researchers, and analysts. It supports SQL and NoSQL databases, allowing users to manage data securely while using AI tools to write queries, fix mistakes, and generate charts and visualizations instantly. Outerbase's table editor, dashboards, and data catalog help users organize, analyze, and share insights efficiently.🔛 Masterclass: AI/LLM TutorialsVoice Trigger System for SiriApple's voice trigger system for Siri includes a first-stage low-power detector to identify potential triggers, and a second-stage, high-precision model to confirm the trigger. It also incorporates speaker identification to ensure the device responds only to its primary user. This sophisticated setup addresses challenges like background noise and phonetically similar words while maintaining power efficiency and privacy.Align Meta Llama 3 to human preferences with DPODPO involves fine-tuning a large language model (LLM) based on feedback from human annotators who rate or rank the model's responses according to desired values, such as helpfulness and honesty. SageMaker Studio provides the computational environment to fine-tune the model using Jupyter notebooks with powerful GPU instances, while SageMaker Ground Truth simplifies the process of gathering human feedback by managing workflows for data annotation. Together, they allow you to align the Llama 3 model’s responses with specific organizational values efficiently.An Intuitive Intro to RLReinforcement learning (RL) is a type of machine learning where an agent learns by interacting with its environment, making decisions, and receiving feedback in the form of rewards or penalties. The goal is to maximize cumulative rewards over time. The agent starts with little to no knowledge and improves through trial and error, learning from past experiences. In RL, actions taken by the agent change the state of the environment, and based on the rewards received, the agent adjusts its future actions. A key concept in RL is balancing exploration (trying new things) and exploitation (using known strategies for rewards).Enhancing LLMs with Structured Outputs and Function CallingEnhancing LLMs with structured outputs and function calling improves their ability to provide accurate and useful responses. Structured outputs ensure consistency and clarity by organizing information in a logical format, reducing ambiguity. Function calling allows LLMs to perform specific tasks, such as retrieving real-time data or executing external functions, making them more interactive and versatile. Combined with techniques like Retrieval-Augmented Generation (RAG), which integrates relevant external information into the model’s responses, these enhancements lead to more reliable, accurate, and contextually rich conversations with LLMs.Safely repairing broken builds with MLGoogle's engineers have developed a machine learning model called DIDACT to automatically repair broken code builds by analyzing historical data of build errors and their fixes. This model suggests potential fixes to developers directly within their Integrated Development Environment (IDE). In a controlled experiment, the use of these machine learning-suggested fixes improved productivity by reducing active coding and feedback time, and increasing the number of completed code changes.🚀 HackHub: AI ToolsAll-Hands-AI/OpenHandsOpenHands is an AI-powered platform designed to assist with software development, allowing agents to perform tasks similar to human developers. These agents can modify code, run commands, browse the web, call APIs, and even use resources like StackOverflow. OpenHands is easy to set up using Docker and can be run in various modes, including scriptable or interactive CLI.langgenius/difyDify is an open-source platform for developing AI applications, offering an intuitive interface that integrates workflows, agent capabilities, model management, and observability features. Dify's core features include a visual AI workflow builder, integration with numerous LLMs, agent tools, and a retrieval-augmented generation (RAG) pipeline for document handling.TransformerOptimus/SuperAGISuperAGI is an open-source framework designed for developers to create, manage, and run autonomous AI agents. It allows seamless operation of multiple agents simultaneously and provides tools to extend their capabilities. With features like graphical interfaces, performance telemetry, and integration with multiple vector databases, SuperAGI enables AI agents to efficiently handle tasks, learn from experience, and optimize token usage.lllyasviel/Paints-UNDOPaints-Undo is an open-source project that provides AI models designed to simulate the drawing process in digital art. By inputting a completed image, users can generate a sequence of steps showing how that image might have been created, mimicking the "undo" function in digital painting software.Stability-AI/stable-audio-toolsStable-Audio-Tools is an open-source library for working with audio generation models. It provides tools for training and running models that generate audio, including a Gradio interface for testing. Users can install the library via PyPI, and the repository includes scripts for both training models and performing inference.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Build a generative AI image description applicationAI_Distilled #75: Rethinking the Role of PPO in RLHF💥 FREE AI & ChatGPT Workshop (Limited time Offer) 🤯An AI-powered professional will earn 10x more. 💰An AI-powered founder will build & scale his company 10x faster 🚀An AI-first company will grow 50x more! 📊🚀Join this 3-hour AI Workshop (worth $399) - FREE for AI_Distilled readers to learn AI strategies & hacks to 10X work output and grow your business.🗓️ Tomorrow | ⏱️ 10 AM ESTWith AI & Chatgpt, you will be able to:✅ Make smarter decisions based on data in seconds using AI✅ Automate daily tasks and increase productivity & creativity✅ Skyrocket your business growth by leveraging the power of AI✅ Save 1000s of dollars by using ChatGPT to simplify complex problems👉 Hurry! Click here to register (FREE for First 100 people only) 🎁SponsoredWelcome to AI_Distilled. Today, we’ll talk about:Awesome AI:Build web applications quickly by generating front-end codePowerful APIs for speech-to-text, text-to-speech, and language understandingv0 by VercelRevolutionize Your Storyboarding ProcessMeasure developer shipping velocity, accuratelyMasterclass:Build a generative AI image description applicationVisualizing and interpreting decision treesRethinking the Role of PPO in RLHFEnhancing Paragraph Generation with a Latent Language Diffusion Model Transparency is often lacking in datasets used to train large language modelsHackHub:A natural language interface for computersLLM app development platform2^x Image Super-ResolutionVideo generation platform based on diffusion modelsPop Audio-based Piano Cover GenerationCheers!Shreyans SinghEditor-in-Chief, Packt🚀 Exclusive for Packt Community: 50% Off Generative AI in Action!Join 25+ top AI experts and access 30+ sessions at our flagship event (Nov 11-13, LIVE). Public tickets are at 35% off, but you get 50% off—our best rate!Limited seats available prices rise by $200 once they're gone. Don’t wait!Book Now with Code BIGSAVE50💻 Awesome AI: Tools for WorkGPT EngineerBuild web applications quickly by generating front-end code using technologies like React, Tailwind, and Vite. Users can describe their app ideas, sync them with GitHub, and deploy them with a single click.OpenHomeAI-powered voice interface that enables natural, seamless conversations with devices using its Voice SDK, allowing any platform to integrate smart voice control. It offers powerful APIs for speech-to-text, text-to-speech, and language understanding, making it ideal for applications like medical transcription and smart home automation. 500 features, including instant translation, emotion detection, and media control.v0 by VercelGenerate web development components and full interfaces quickly using chat-based prompts. It helps developers create UI elements like buttons, modals, and pages by simply describing what they need, enabling faster development workflows.StoryboarderRapidly transform ideas into detailed storyboards, animatics, and screenplays. With features like Image-To-Video, the platform can turn static images into dynamic videos, enhancing storytelling and saving time. It supports various media projects, including commercials, films, and social media content, and offers integrated scriptwriting, consistent art styles, and expert support to streamline the creative process.Maxium AIAccurately measure developer efficiency by tracking shipping velocity and performance, going beyond just lines of code or commits. It integrates with GitHub to provide a standardized evaluation mechanism across different tech stacks and programming languages.🔛 Masterclass: AI/LLM TutorialsBuild a generative AI image description applicationThis guide explains how to build an application for generating image descriptions using Anthropic's Claude 3.5 Sonnet model on Amazon Bedrock and AWS CDK. By integrating Amazon Bedrock’s multimodal models with AWS services like Lambda, AppSync, and Step Functions, you can quickly develop a solution that processes images and generates descriptions in multiple languages. The use of Generative AI CDK Constructs streamlines infrastructure setup, making it easier to deploy and manage the application.Visualizing and interpreting decision treesTensorFlow recently introduced a tutorial on using dtreeviz, a leading visualization tool, to help users visualize and interpret decision trees. dtreeviz shows how decision nodes split features and how training data is distributed across different leaves. For example, a decision tree might use features like the number of legs and eyes to classify animals. By visualizing the tree with dtreeviz, you can see how each feature influences the model's predictions and understand why a particular decision was made.Rethinking the Role of PPO in RLHFIn Reinforcement Learning with Human Feedback (RLHF), there's a challenge where the reward model uses comparative feedback (i.e., comparing multiple responses) while the fine-tuning phase of RL uses absolute rewards (i.e., evaluating responses individually). This discrepancy can lead to issues in training. To address this, researchers introduced Pairwise Proximal Policy Optimization (P3O), a new method that integrates comparative feedback throughout the RL process. By using a pairwise policy gradient, P3O aligns the reward modeling and fine-tuning stages, improving the consistency and effectiveness of training. This approach has shown better performance in terms of reward and alignment with human preferences compared to previous methods.Enhancing Paragraph Generation with a Latent Language Diffusion Model The PLANNER model, introduced in 2023, enhances paragraph generation by combining latent semantic diffusion with autoregressive techniques. Traditional models like GPT often produce repetitive or low-quality text due to "exposure bias," where the training and inference processes differ. PLANNER addresses this by using a latent diffusion approach that refines text iteratively, improving coherence and diversity. It encodes paragraphs into latent codes, processes them through a diffusion model, and then decodes them into high-quality text. This method reduces repetition and enhances text quality.Transparency is often lacking in datasets used to train large language modelsA recent study highlights the lack of transparency in datasets used to train large language models (LLMs). As these datasets are combined from various sources, crucial information about their origins and usage restrictions often gets lost. This issue not only raises legal and ethical concerns but can also impact model performance by introducing biases or errors if the data is miscategorized. To address this, researchers developed the Data Provenance Explorer, a tool that provides clear summaries of a dataset’s origins, licenses, and usage rights.🚀 HackHub: AI ToolsOpenInterpreter/open-interpreterOpen Interpreter is a tool that allows language models (like GPT-4) to execute code locally on your machine, supporting languages like Python, JavaScript, and shell scripts. It works like ChatGPT but with the ability to interact with your system's resources.langgenius/difyDify is an open-source platform for developing AI applications using large language models (LLMs). It provides an intuitive interface for building AI workflows, managing models, and integrating tools like Google Search or DALL·E. Dify supports a wide variety of LLMs and offers features like a prompt IDE, document retrieval (RAG), agent-based automation, and detailed observability for monitoring performance.Tohrusky/Final2xFinal2x is a cross-platform tool designed to enhance image resolution and quality using advanced super-resolution models such as RealCUGAN, RealESRGAN, and Waifu2x. It's ideal for anyone looking to improve image resolution efficiently across various platforms.ali-vilab/VGenVGen is an open-source video generation platform from Alibaba's Tongyi Lab that offers a wide range of tools for generating videos from various inputs like text, images, and motion instructions. It features state-of-the-art models like I2VGen-xl for image-to-video synthesis and DreamVideo for custom subject and motion generation. VGen supports tasks like video generation from human feedback and video latent consistency modeling.sweetcocoa/pop2pianoPop2Piano is a deep learning model that automatically generates piano covers from pop music audio. Traditionally, creating a piano cover involves understanding the song's melody, chords, and mood, which is challenging even for humans. Prior methods used melody and chord extraction, but Pop2Piano skips these steps, directly converting pop music waveforms into piano covers using a Transformer-based approach. The model was trained on a large dataset of synchronized pop songs and piano covers (300 hours), enabling it to generate plausible piano performances without explicit musical extraction modules.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Transform your database into your AI platformAI_Distilled #74: Unlocking 7B+ language models in your browser: A deep dive with Google AI Edge's MediaPipe200+ hours of research on AI tools & hacks packed in 3 hoursThis free 3-hour Training on AI & ChatGPT (worth $399) will help you become a master of 20+ AI tools & prompting techniques and save 16 hours/week.Get it now for absolutely free! (for first 100 users only) 🎁You will learn how to:- Build business that make $10,000 by just using AI tools- Make quick & smarter decisions using AI-led data insights- Write emails, content & more in seconds using AI- Solve complex problems, research 10x faster & save 16 hours every weekRegister & save your seat now! (100 free seats only)SponsoredWelcome to AI_Distilled. Today, we’ll talk about:Awesome AI:LM Studio - Discover, download, and run local LLMsPainless Data Extraction and Web AutomationFleak AI Serverless API BuilderListen to Actual Clients' FeedbackTheysaid - Conversational AI SurveysMasterclass:Unlocking 7B+ language models in your browser: A deep dive with Google AI Edge's MediaPipeDeploying Attention-Based Vision Transformers to Apple Neural EngineMistral-NeMo: 4.1x Smaller with Quantized MinitronConnect the Amazon Q Business generative AI coding companion to your GitHub repositoriesAugmenting recommendation systems with LLMsHackHub:high-performance, multiplayer code editor from the creators of Atom and Tree-sitter.Multi-Platform Package Manager for Stable DiffusionSharpen your low-resolution pictures with the power of AI upscalingTransform your database into your AI platformLarge language model series developed by Qwen team, Alibaba Cloud.Cheers!Shreyans SinghEditor-in-Chief, Packt💻 Awesome AI: Tools for WorkLM Studio - Discover, download, and run local LLMsLM Studio 0.3.0 is a major update to the local LLM desktop application that enhances its offline capabilities with new features. Users can now chat with documents, using either full document context or "Retrieval Augmented Generation" (RAG) for longer texts. The update also introduces an OpenAI-like JSON output API, customizable UI themes, and automatic hardware detection for optimal performance.Painless Data Extraction and Web Automation (agentql.com)AgentQL is a powerful tool for data extraction and web automation that uses AI to reliably find and interact with web elements, even as websites change. Unlike traditional methods that rely on fragile XPath or DOM selectors, AgentQL allows users to locate elements using natural language descriptions, making it easier to automate tasks like filling forms, gathering data, and conducting end-to-end testing.Fleak AI Workflows. Simplified | Serverless API Builder | fleak.aiFleak is a low-code, serverless API builder designed for data teams to quickly and easily create, integrate, and scale AI and data workflows without managing any infrastructure. It allows users to configure and deploy workflows in minutes, seamlessly integrating with tools like large language models, vector databases, and modern storage technologies.Listen to Actual Clients' Feedback | Seven24 AISeven24 helps you capture and act on user feedback with ease. Integrate their tool into your product to collect feedback via text or voice, and their AI transforms this feedback into actionable tasks. With features like sentiment analysis, you can boost positive reviews and address issues quickly.Theysaid - Conversational AI SurveysTheySaid offers the world’s first conversational AI survey, designed to significantly increase response rates and improve customer engagement. By integrating seamlessly with your existing tech stack, the AI tool generates personalized survey questions based on your website content and follows up with users through conversational interactions.🔛 Masterclass: AI/LLM TutorialsUnlocking 7B+ language models in your browser: A deep dive with Google AI Edge's MediaPipeGoogle AI Edge's MediaPipe has developed a new system that allows large language models (LLMs) to run directly in web browsers, overcoming memory and performance limitations. By using WebAssembly and WebGPU, MediaPipe can now load and execute models like Gemma 1.1 with 7 billion parameters, which was previously unfeasible in-browser. The approach includes breaking down models into manageable parts and leveraging efficient memory usage techniques to handle the massive size of LLMs.Deploying Attention-Based Vision Transformers to Apple Neural EngineThe concept of Vision Transformers (ViTs) was introduced to leverage transformer models, which were originally used in natural language processing, for image recognition tasks. Unlike traditional Convolutional Neural Networks (CNNs), Vision Transformers process images by dividing them into smaller patches and applying attention mechanisms. This approach can handle various computer vision tasks such as image classification and object detection more effectively.Mistral-NeMo: 4.1x Smaller with Quantized MinitronNVIDIA's Minitron technique makes large language models (LLMs) like Mistral-NeMo smaller and more efficient by removing less critical parts and retraining them. This process reduces the models' sizes while keeping their performance high. The Minitron version of Mistral-NeMo, for instance, shrinks the model from 12 billion to 8 billion parameters. Combining Minitron with 4-bit quantization further compresses these models, allowing them to run on smaller GPUs and reducing operational costs.Connect the Amazon Q Business generative AI coding companion to your GitHub repositoriesYou can link Amazon Q Business, an AI-powered assistant, to your GitHub repositories using the Amazon Q GitHub (Cloud) connector. This setup allows you to use natural language queries to access information like commits, issues, and pull requests from your GitHub repositories. By integrating this tool, your development team can boost productivity, reduce context switching, and quickly retrieve information from your GitHub data through a conversational interface.Augmenting recommendation systems with LLMsLarge language models (LLMs), like Google's PaLM, can significantly enhance recommendation systems by integrating advanced AI capabilities. By incorporating LLMs into the recommendation pipeline, you can improve features like conversational recommendations, sequential recommendations based on user activity, and rating predictions. LLMs can interactively suggest items, understand the sequence of user preferences, and predict ratings with high accuracy.🚀 HackHub: AI Toolszed-industries/zedZed is a high-performance, multiplayer code editor developed by the team behind Atom and Tree-sitter. It can be installed on macOS and Linux directly or through package managers, though it’s not yet available for Windows or web platforms.LykosAI/StabilityMatrixStability Matrix is a multi-platform tool designed for managing Stable Diffusion Web UI packages across Windows, Linux, and macOS. It features a customizable interface with a syntax-highlighted terminal, a model browser for importing models from CivitAI and HuggingFace, and a shared model directory for all packages.Lucchetto/SuperImageSuperImage is an Android app that uses AI to enhance low-resolution images by upscaling them to higher resolutions. Built with the MNN framework and Real-ESRGAN, it processes images in tiles on the device's GPU, merging them into a high-resolution final image. It requires Android 7 or above and support for Vulkan or OpenCL.superduper-io/superduperIntegrate AI models and machine learning workflows with your database to implement custom AI applications, without moving your data. Including streaming inference, scalable model hosting, training and vector search.QwenLM/Qwen2Qwen2 is a suite of advanced language models available in various sizes, including up to 72 billion parameters. It offers state-of-the-art performance in tasks like coding and math, and supports up to 128K tokens for extended context. The models are pretrained and instruction-tuned, and they are available for use through Hugging Face and ModelScope.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

xAI, Elon Musk's AI startup, launches an API AI_Distilled #73: Introducing the new Claude 3.5 Sonnet, and Claude 3.5 Haiku and “Computer Use” 🚀 The Most Awaited 2-for-1 Deal Drops Tomorrow! 🚀 Unlock our 2-for-1 offer at Generative AI in Action (Nov 11-13) and bring a friend, colleague, or your team to double the learning experience. 🗓 Sale Starts: Tomorrow, Friday, Oct 25, 10 AM ET ⏳ Duration: 24 hours only Don’t miss out—mark your calendar and get ready to grab this exclusive deal! Join 25+ AI Experts, 30+ Sessions & 1000+ Tech Pros Welcome to AI_Distilled. Today, we’ll talk about: Techwave: xAI, Elon Musk's AI startup, launches an API Introducing Stable Diffusion 3.5 Introducing the new Claude 3.5 Sonnet, and Claude 3.5 Haiku and “Computer Use” Meta releases Spirit LM, open-source multimodal modelintegrating text and speech seamlessly New autonomous agents scale your team like never before Awesome AI: guidde・Magically create video documentation with AI Feta - Better stand-ups, retros, syncs and more BrowserCopilot AI - Your AI Companion Across the Web MyLens.ai: Key Points of any Webpage & Youtube with one click Trag: Superlinter for any stack Masterclass: Solving complex problems with OpenAI o1 models Thinking LLMs:General Instruction Following with Thought Generation Agent-as-a-Judge: Evaluate Agents with Agents Learn dynamic few-shot prompting with LlamaIndexworkflows for enhanced LLM performance Fine-tuning LLMs to 1.58-bit: compress models without sacrificing performance HackHub 3b1b/videos: Code for the manim-generated scenes used in 3blue1brown videos phidatahq/phidata: Build AI Agents with memory, knowledge, tools and reasoning. Chat with them using a beautiful Agent UI. ComposioHQ/composio: Composio equip's your AI agents & LLMs with 100+ high-quality integrations via function calling Janus: Any-to-Anyautoregressive frameworkfor multimodal AI. Ichigo: Llama learns to talk - Homebrew Cheers! Shreyans Singh Editor-in-Chief, Packt ⚡ TechWave: AI/GPT News & Analysis xAI, Elon Musk's AI startup, launches an API Elon Musk’s AI startup, xAI, has launched an API for its generative AI model, Grok, allowing developers to integrate Grok’s features into their applications. The API currently offers a single model, "grok-beta," priced at $5 per million input tokens and $15 per million output tokens. Grok, which powers various features on X (formerly Twitter), is known for its rebellious, uncensored responses and image generation capabilities. Although still developing, xAI aims to catch up to competitors like OpenAI and Anthropic, using data from Musk's companies and X to train future models. Introducing Stable Diffusion 3.5 Stable Diffusion 3.5 is the latest release from Stability AI, offering multiple highly customizable models designed to run on consumer hardware. These models, including Stable Diffusion 3.5 Large and Large Turbo, are available for free for most uses under a permissive license. They offer a balance of high image quality, fast performance, and flexibility, making them ideal for creators, researchers, and businesses. The models can generate diverse images in various styles and are available for download on platforms like Hugging Face. Introducing the new Claude 3.5 Sonnet, and Claude 3.5 Haiku and “Computer Use” Anthropic has announced updates to its Claude 3.5 models, including the upgraded Claude 3.5 Sonnet, which excels in coding and tool use, and the new Claude 3.5 Haiku, which offers similar performance to previous top-tier models at a lower cost and faster speed. They’ve also introduced a groundbreaking “computer use” capability in public beta, allowing Claude to interact with computers like a human by navigating interfaces, clicking buttons, and typing. This feature is still experimental but has potential for automating complex tasks. Meta releases Spirit LM, open-source multimodal modelintegrating text and speech seamlessly Meta has released Spirit LM, a model for handling both spoken and written language in an interleaved manner. The repository contains model weights, inference code, and evaluation scripts for the Spirit LM model, which can be set up using Conda or pip. It includes tools for speech tokenization and text generation, with an emphasis on preserving speech-text sentiment in its outputs. New autonomous agents scale your team like never before Microsoft announced new autonomous agent capabilities in Copilot Studio to help businesses scale more efficiently. Starting next month, businesses will be able to create their own agents, designed to handle tasks like sales, supply chain management, and customer service. These agents, integrated into Dynamics 365, can automate complex processes such as lead generation, supplier communication, and customer support. 💻 Awesome AI: Tools for Work guidde・Magically create video documentation with AI Guidde is an AI-powered platform designed to help businesses quickly create video documentation, making complex workflows easier to explain. It enables users to capture processes using a browser extension or desktop app and automatically generates step-by-step instructions with customizable AI-generated voiceovers. Feta - Better stand-ups, retros, syncs and more Feta is a platform designed to help product and engineering teams run more efficient meetings by streamlining tasks and capturing key insights. It auto-compiles updates for standups, integrates with tools like Jira and GitHub, and generates actionable meeting summaries and notes. BrowserCopilot AI - Your AI Companion Across the Web Yaseen AI is a browser-based AI companion that helps professionals work more efficiently by providing real-time assistance on any website. It integrates seamlessly with workflows, offering personalized responses and support through its Copilot feature. MyLens.ai: Key Points of any Webpage & Youtube with one click MyLens.ai is a Chrome extension that transforms any webpage or YouTube video into visual summaries like mindmaps, timelines, tables, and flowcharts with just one click. It helps users quickly extract key insights from long articles, reports, or videos, saving time by breaking down complex content into clear, shareable visuals. Trag: Superlinter for any stack Superlinter, powered by Trag, is a versatile tool that allows developers to replace traditional linters and code analysis tools with a natural language-based linter that works for any programming language. Users can describe specific code patterns or rules in plain English, which the linter then enforces within their code. 🔛 Masterclass: AI/LLM Tutorials Solving complex problems with OpenAI o1 models Thinking LLMs:General Instruction Following with Thought Generation Large Language Models are typically trained to respond to user instructions based on patterns in data, but they lack the ability to think explicitly before answering. This is important for complex tasks that require reasoning or planning. To address this, a method called Thought Preference Optimization (TPO) allows LLMs to develop thinking abilities without additional human data. The process involves generating multiple potential thoughts, evaluating the quality of the final responses, and optimizing them through reinforcement learning. Agent-as-a-Judge: Evaluate Agents with Agents The "Agent-as-a-Judge" framework is a new method for evaluating agentic systems, where agents are used to evaluate other agents instead of relying on human evaluators or traditional methods that only consider final outcomes. This framework provides feedback throughout the task-solving process, which is important for agentic systems that act step-by-step, like humans. Applied to code generation, "Agent-as-a-Judge" proved more effective and reliable than the existing LLM-as-a-Judge framework and performed similarly to human evaluators, but at a much lower cost and time. Learn dynamic few-shot prompting with LlamaIndexworkflows for enhanced LLM performance In LlamaIndex, workflows are event-driven systems where functions are chained together as steps, each handling specific event types. By using the `@step` decorator, the system ensures that steps only run when a valid event is received, and each step can emit new events for the next. Workflows enable creating processes like agents, document extraction, or retrieval-augmented generation (RAG) pipelines. They are fully asynchronous, allowing efficient parallel processing, and come with built-in observability. Users can integrate global contexts, handle multiple events, and even retry steps in case of failures. Fine-tuning LLMs to 1.58-bit: compress models without sacrificing performance Fine-tuning large language models (LLMs) to use only 1.58 bits per parameter (based on the BitNet architecture) dramatically reduces their computational and memory requirements by using extreme quantization. This process limits the values of each parameter to just three options: -1, 0, and 1. Although such quantization typically requires training a model from scratch, the authors have found ways to fine-tune pre-trained models to achieve similar efficiency without losing significant performance. 🚀 HackHub: AI Tools 3b1b/videos: Code for the manim-generated scenes used in 3blue1brown videos This project contains the code used to create the math videos by 3Blue1Brown, primarily using the Manim library, a tool for generating mathematical animations. While the Manim library itself is open source under the MIT license, the content in this repository is under a Creative Commons license (CC BY-NC-SA 4.0), which allows sharing and adapting with credit but not for commercial purposes. phidatahq/phidata: Build AI Agents with memory, knowledge, tools and reasoning. Chat with them using a beautiful Agent UI. Phidata is a framework for building intelligent agents equipped with memory, knowledge, tools, and reasoning capabilities. You can create agents for various tasks, like web search or financial data analysis, and even combine them into teams to work together. ComposioHQ/composio: Composio equip's your AI agents & LLMs with 100+ high-quality integrations via function calling Composio is a toolset that helps developers build AI agents equipped with a wide range of pre-configured tools and integrations with minimal effort. It simplifies tasks like authentication, accuracy, and reliability, enabling developers to create agents that can interact with platforms like GitHub, Notion, Slack, and more. Janus: Any-to-Anyautoregressive frameworkfor multimodal AI. Janus is an advanced multimodal framework that improves the way AI models understand and generate both visual and textual content. It separates the visual encoding process into distinct pathways but maintains a unified transformer architecture, which increases flexibility and performance for various tasks. Ichigo: Llama learns to talk - Homebrew Ichigo is a new speech and text multimodal model built on Llama3-s, designed for understanding and generating both audio and text. Developed through open research by the Homebrew Computer Company, Ichigo addresses key limitations in earlier models, such as limited multilingual capabilities and issues with recognizing nonspeech inputs. 📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us. If you have any comments or feedback, just reply back to this email. Thanks for reading and have a great day! *{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{line-height:0;font-size:75%} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Books on AI handpicked for youAre you ready to enhance your expertise and stay ahead of the curve in the latest tech trends? Dive into cutting-edge resources designed to elevate your skills.Whether you're exploring AI, refining your techniques, or mastering AI, we have the perfect reads for you.BESTSELLERS OF THE WEEKBuilding LLM Powered ApplicationsBy Valentina AltoEmbed LLMs into real-world applicationsUse LangChain to orchestrate LLMs and their components within applicationsGrasp basic and advanced techniques of prompt engineeringeBook: $19.99 $39.99Print: $34.98 $49.99Building Data-Driven Applications with LlamaIndexBy Andrei GheorghiuExamine text chunking effects on RAG workflows and understand security in RAG app developmentDiscover chatbots and agents and learn how to build complex conversation enginesBuild as you learn by applying the knowledge you gain to a hands-on projecteBook: $24.99 $35.99Print: $30.99 $44.99Deep Learning with TensorFlow and KerasBy Amita Kapoor, Antonio Gulli, Sujit PalUnderstand the fundamentals of deep learning and machine learning through clear explanations and extensive code samplesImplement graph neural networks, transformers using Hugging Face and TensorFlow Hub, and joint and contrastive learningLearn cutting-edge machine and deep learning techniqueseBook: $27.98 $39.99Print: $44.99Data Modeling with SnowflakeBy Serge GershkovichLearn core modeling techniques tied to practical examples using native Snowflake architectureAdopt a universal modeling language to communicate business value to functional teamsGo beyond physical modeling with SQL recipes to transform and shape your Snowflake dataeBook: $27.98 $39.99Print: $39.98 $49.99Databricks ML in ActionBy Stephanie Rivera, Anastasia Prokaieva, Amanda Baker, Hayley HornBuild machine learning solutions faster than peers only using documentationEnhance or refine your expertise with tribal knowledge and concise explanationsFollow along with code projects provided in GitHub to accelerate your projectseBook: $24.99 $35.99Print: $39.99 $44.99Want even more resources? Start a free trial and explore our entire library! From cloud solutions to system programming, gain unlimited access to the latest in tech. Start your free trial today.DISCOVER TRENDING TITLESThanks,PacktCopyright (C) 2024 Packt Publishing. All rights reserved.Our mailing address is:Packt Publishing, Grosvenor House,11 St Paul's Square, Birmingham,West Midlands, B3 1RB, United KingdomWant to change how you receive these emails?You can update your preferences or unsubscribe*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{line-height:0;font-size:75%}#converted-body .list_block ol,#converted-body .list_block ul,.body [class~=x_list_block] ol,.body [class~=x_list_block] ul,u+.body .list_block ol,u+.body .list_block ul{padding-left:20px} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Make charts on Perplexity code interpreterAI_Distilled #72: Mistral AI Launches Ministral 3B and 8B Models for On-Device AI ComputingJoinGenerativeAI InActionnow withaFull Event Pass for just $239.99—40% off the regular price—with codeFLASH40.BOOK TODAY AT $239.99 $399.99Three Reasons Why You Cannot Miss This Event:Network with 25+ Leading AI ExpertsGain Insights from 30+ Dynamic Talks and Hands-On SessionsEngage with Experts and Peers through 1:1 Networking, Roundtables, and AMAsAct fast—this FLASH SALE is only for a limited number of seats!BOOK TODAY AT $239.99 $399.99Welcome to AI_Distilled. Today, we’ll talk about:Techwave:Mistral AI Launches Ministral 3B and 8B Models for On-Device AI ComputingMake charts on perplexity code interpreterIntroducing Swarm: OpenAI’s New Open-Source Multi-Agent Orchestration FrameworkOpenAI MLE-bench: Evaluating Machine Learning Agents on Machine Learning EngineeringAnthropic’s Responsible Scaling Policy, October 15, 2024Awesome AI:MU - Perplexity FinanceAdobe Launches Firefly Video Model and Enhances Image, Vector and Design ModelsYou can now search with Google Lens in the Chromebook Gallery appGradioStrella - AI-Powered Customer ResearchMasterclass:Aria: First Open Multimodal Native MOE ModelUnderstanding the Limitations of Mathematical Reasoning in Large Language ModelsNo Priors Ep. 80 | With Andrej Karpathy from OpenAI and TeslaMulti document agentic RAG: A walkthroughLLMs From Scratch Ch05/08:_Memory efficient_weight_loadingHackHubLlama-3.1-Nemotron-70B - a nvidia Collectionmlc-ai/mlc-llm: Universal LLM Deployment Engine with ML CompilationSurya: OCR, layout analysis, reading order, table recognition in 90+ languagesTEN-Agent: world’s first real-time multimodal agent integrated with the OpenAI Realtime APICinnamon/kotaemon: An open-source RAG-based tool for chatting with your documentsCheers!Shreyans SinghEditor-in-Chief, PacktLooking to build, train, deploy, or implement Generative AI?Meet Innodata — offering high-quality solutions for developing and implementing industry-leading generative AI.With 5,000+ in-house SMEs and expansion and localization supported across 85+ languages, Innodata drives AI initiatives for enterprises globally.Learn More⚡ TechWave: AI/GPT News & AnalysisMistral AI Launches Ministral 3B and 8B Models for On-Device AI ComputingMistral AI has introduced two new advanced models, Ministral 3B and Ministral 8B, designed for efficient on-device and edge computing. These models, which are more powerful and faster than their predecessors, excel in areas like knowledge, reasoning, and task execution, making them ideal for privacy-focused, offline applications such as local translation and robotics. With a large context length and specialized attention patterns, they offer low-latency and cost-effective solutions for a variety of uses, from personal projects to industrial tasks. Both models are now available for commercial and research use.Make charts on perplexity code interpreterIntroducing Swarm: OpenAI’s New Open-Source Multi-Agent Orchestration FrameworkSwarm is an experimental, educational framework developed by OpenAI to explore lightweight orchestration of multiple agents in a flexible and ergonomic way. It allows developers to create and manage multi-agent systems where agents can pass tasks or conversations between each other, handling complex workflows efficiently. Designed for educational purposes, Swarm uses OpenAI’s Chat Completions API, with agents executing Python functions and handling different tasks.OpenAI MLE-bench: Evaluating Machine Learning Agents on Machine Learning EngineeringMLE-bench is a benchmark created by OpenAI to evaluate how well AI agents can perform tasks related to machine learning engineering. It uses 75 competitions from Kaggle to test real-world skills such as training models, preparing datasets, and running experiments. Human baselines are established using Kaggle's leaderboards, and the best-performing AI setup, OpenAI's o1-preview with AIDE scaffolding, achieves results comparable to a Kaggle bronze medal in about 17% of competitions.Anthropic’s Responsible Scaling Policy, October 15, 2024Anthropic's updated Responsible Scaling Policy (RSP) outlines its commitment to ensuring that AI models do not cause catastrophic harm by implementing safety and security measures. The policy introduces AI Safety Level (ASL) Standards, which become stricter as AI capabilities increase. These standards help determine when models need stronger safeguards. The update includes guidelines for assessing models based on Capability Thresholds, focusing on areas like chemical, biological, radiological, and nuclear (CBRN) risks. If a model reaches a higher capability, additional safeguards (ASL-3 or higher) are required to mitigate risks.💻 Awesome AI: Tools for WorkMU - Perplexity FinancePerplexity revealed a preview of its upcoming financial analysis platform, "Perplexity for Finance," designed to provide users with real-time stock quotes, historical earnings reports, industry comparisons, and detailed financial data, all through an intuitive and user-friendly interface. A video shared by the company demonstrated how users can easily access and visualize financial data, such as Nvidia’s earnings history and stock price trends.Adobe Launches Firefly Video Model and Enhances Image, Vector and Design ModelsAdobe has launched its new Firefly Video Model (beta), expanding its AI-powered creative tools to video content, marking the first such model designed for safe commercial use. In addition to this, Adobe enhanced its Firefly Image, Vector, and Design models, offering faster image generation and new capabilities integrated into apps like Photoshop, Illustrator, and Premiere Pro. These tools allow users to generate videos and images from text prompts, extend video clips, and more.You can now search with Google Lens in the Chromebook Gallery appChromebooks now have Google Lens integrated into their Gallery app, allowing users to quickly search for information related to any image or document they view. By opening a file in the Gallery app, users can select a section of the image or document and use Google Lens to perform a search. This new feature acts as a shortcut to Chrome’s existing Google Lens tool, saving users time by streamlining the process of capturing and searching with images.GradioGradio 5.0 is a user-friendly tool that makes it easy to create web-based interfaces for machine learning models. With just a few lines of Python code, developers can build interactive apps that allow anyone to test and interact with their models. Gradio can be embedded in notebooks or shared via public links, and it supports integration with various Python libraries. It also offers permanent hosting on Hugging Face Spaces. Gradio is widely used by companies like Google and Amazon, as well as researchers and developers for quick and efficient model demos.Strella - AI-Powered Customer ResearchStrella is an AI-powered tool designed to streamline customer research by automating interviews, recruitment, and analysis. It helps researchers quickly create custom interview guides, conduct AI-moderated interviews, and analyze insights in real-time, making decisions faster and more informed. Strella handles logistics like scheduling and incentives, allowing researchers to focus on higher-impact tasks. It supports global participants, runs interviews 24/7, and offers features like dynamic follow-up questions, screen recording, and multilingual capabilities. The platform boosts efficiency, speeds up research timelines, and enhances research output.🔛 Masterclass: AI/LLM TutorialsAria: First Open Multimodal Native MOE ModelRhymes AI introduced Aria, an open-source multimodal native Mixture-of-Experts (MoE) model, designed to process various input types—text, images, video, and code—simultaneously. It excels in tasks involving complex multimodal data and offers a long context window of up to 64,000 tokens, making it highly efficient for tasks like video captioning or document understanding. Aria outperforms other open and some proprietary models like GPT-4o and Gemini-1.5, demonstrating competitive performance with fewer activated parameters.Understanding the Limitations of Mathematical Reasoning in Large Language ModelsRecent advancements in Large Language Models (LLMs) have led to interest in their ability to handle formal reasoning, especially in math. The widely used GSM8K benchmark tests models on grade-school-level math questions, but it's unclear if improvements in scores reflect true advances in reasoning. To address this, researchers created GSM-Symbolic, a new benchmark with symbolic templates that generate more varied and controlled questions. They found that LLMs struggle when numerical values or clauses are slightly changed in questions, suggesting that current models rely on patterns from training data rather than genuine logical reasoning.No Priors Ep. 80 | With Andrej Karpathy from OpenAI and TeslaIn this episode of the *No Priors* podcast, Andrej Karpathy, a key figure in AI and former leader of Tesla Autopilot, discusses the evolution of self-driving cars, comparing Tesla's approach with Waymo's. He also touches on Tesla's Optimus humanoid robot and the challenges in robotics and AI today. Karpathy explores the potential for integrating AI with human cognition and shares insights on AI-driven education and its impact on future learning. He also talks about his new venture, Eureka Labs, and offers advice on what young people should study to prepare for a future shaped by AI advancements.Multi document agentic RAG: A walkthroughThis blog post by Vipul Maheshwari explains the concept of Agentic Retrieval-Augmented Generation (RAG), an advanced version of traditional RAG systems. Unlike basic RAG models that retrieve relevant data for language models to generate responses, Agentic RAG introduces decision-making autonomy. It can analyze a task, break it into smaller steps, and take actions without constant supervision. The post walks through how to build an Agentic RAG system for car diagnostics using LanceDB, LlamaIndex, and vector databases.LLMs From Scratch Ch05/08:_Memory efficient_weight_loading🚀 HackHub: AI ToolsLlama-3.1-Nemotron-70B - a nvidia CollectionNVIDIA has released several advanced AI models on Hugging Face, including the Llama-3.1-Nemotron series, which offers state-of-the-art (SOTA) performance on benchmarks like Arena Hard and RewardBench. These models, like Llama-3.1-Nemotron-70B, focus on text generation and include variations tailored for instruction-following (Instruct) and reward-based tasks. NVIDIA's collection also includes models for specialized tasks such as speech synthesis (Parakeet) and reinforcement learning with human feedback (RLHF).mlc-ai/mlc-llm: Universal LLM Deployment Engine with ML CompilationMLC LLM is an open-source project that provides a universal deployment engine for large language models (LLMs) with machine learning compilation (MLC). Its goal is to enable developers to optimize and deploy AI models across various platforms, such as AMD, NVIDIA, and Apple GPUs, and even on mobile devices like iOS and Android.Surya: OCR, layout analysis, reading order, table recognition in 90+ languagesSurya is an open-source document OCR (Optical Character Recognition) toolkit that supports over 90 languages. It offers advanced features like text detection, layout analysis (including tables, images, and headers), reading order detection, and table recognition, working efficiently across a wide range of documents, from scientific papers to forms.TEN-Agent: world’s first real-time multimodal agent integrated with the OpenAI Realtime APITEN Agent is a real-time multimodal AI agent that integrates the OpenAI Realtime API and RTC for ultra-low latency performance. The agent can be extended with edge-cloud integrations, real-time state management, and drag-and-drop tools for complex applications.Cinnamon/kotaemon: An open-source RAG-based tool for chatting with your documentsKotaemon is an open-source tool designed for interacting with documents through a Question Answering (QA) system built on Retrieval-Augmented Generation (RAG) technology. It supports various large language models (LLMs), both local and via APIs (like OpenAI), and allows users to ask questions about their documents.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{line-height:0;font-size:75%} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}