AI_Distilled 37: Cutting-Edge Updates and Expert Guidance

👋 Hello ,

“[Sovereign AI] codifies your culture, your society's intelligence, your common sense, your history – you own your own data…[Use AI to] activate your industry, build the infrastructure, as fast as you can.” - Jensen Huang, NVIDIA founder and CEO 

Huang strongly advocated for countries to rapidly develop their own national AI capabilities and systems. NVIDIA's dominance in GPUs positions it to be a major beneficiary of the AI revolution as its technologies are fundamental for running advanced AI applications. No wonder NVIDIA's market value recently surpassed Amazon’s. 

Embark on a new AI journey with AI_Distilled, a curated digest of the most recent developments in AI/ML, LLMs, NLP, GPT, and Gen AI. We’ll kick things off by tapping into the latest news and developments in the AI sector: 

OpenAI updates ChatGPT with memory retention 

Microsoft unveils new Copilot features 

Google updates Gemini and unveils mobile app 

Apple's new AI model called MGIE 

New open-source AI model Aya converses in 100+ languages 

Reka AI introduces two new state-of-the-art AI models 

DeepMind and USC develop new technique to improve LLMs’ reasoning abilities 

New open-source AI model Smaug-72B achieves top spot 

NVIDIA unveils new chatbot Chat with RTX 

AI helps identify birds and conserve an English wetland 

USPTO issues new guidance stating AI alone can't be named as an inventor 

We’ve also handpicked GPT and LLM resources and secret knowledge that’ll come in handy for your next project:  

Building a Scalable Foundation Model Platform for Your Enterprise 

Making Bridges Between AI and Business 

Evaluating Large Language Models: A Guide to Benchmark Tests 

Code Generation Gets Smarter with Context 

Looking for hands-on tips and strategies straight from the developer community? We’ve got you covered with some incredible tutorials to get you started: 

Building a Question Answering Bot from Scratch 

Creating SMS Apps with Next.js and AI Assistants 

Harness the Power of LLMs Without GPUs 

Making the Switch to Open-Source Models 

Finally, feel free to check out our curated list of smoking hot GitHub repositories. 

arplaboratory/learning-to-fly 

time-series-foundation-models/lag-llama 

noahfarr/rlx 

uclaml/SPIN 

phidatahq/phidata 

📥 Feedback on the Weekly Edition

Take our weekly survey and get a free PDF copy of our best-selling book, "Interactive Data Visualization with Python - Second Edition." 

📣 And here's the twist – we're tuning into YOUR frequency! Inspired by a reader's request, we're launching a column just for you. Got a burning question or a topic you're itching to dive into? Drop your suggestions in our content box – because your journey of discovery is our blueprint.

We appreciate your input and hope you enjoy the book! 

Writer’s Credit: Special shout-out to Vidhu Jain for their valuable contribution to this week’s newsletter content!  

Cheers,  

Merlyn Shelley  

Editor-in-Chief, Packt 

SignUp | Advertise | Archives

⚡ TechWave: AI/GPT News & Analysis

OpenAI (whose annual revenue hit $2 billion in December) has updated its ChatGPT chatbot to retain information from past conversations, allowing it to apply context from previous discussions to new chats. This could help the bot respond with more relevant, personalized replies. Microsoft's AI assistant Copilot has also received upgrades like improved models and image editing tools. Google updated its conversational AI too with a new mobile app and advanced model (now called Gemini). A new paid tier for Gemini Ultra provides developers and users access to more advanced features and capabilities for $20 per month.  

Courtesy: OpenAI 

Apple's new AI model called MGIE allows editing images through natural language commands, performing tasks from color adjustments to complex manipulations. Interestingly, a new open-source AI model called Aya can converse in over 100 languages, potentially increasing access for many. Reka AI has introduced two new state-of-the-art AI models, Flash and Edge, which achieve top performance on language and vision tasks while Edge maintains a smaller size. Researchers from DeepMind and USC have developed a new technique called SELF-DISCOVER to improve LLMs’ reasoning abilities. A new open-source AI model (available for all) called Smaug-72B has achieved the top spot on the leaderboard for language models, demonstrating skills that surpass proprietary competitors.

Courtesy: Apple 

NVIDIA's market value surpassed Amazon's for the first time since 2002 thanks to strong demand for its AI chips. The company also released a new chatbot called Chat with RTX allowing users to run personalized generative AI models locally on PCs. Chatbots aside, AI is making waves across fields. Conservationists are using AI to identify birds by their songs to help restore an English wetland. Scientists are speeding discoveries and tackling climate change better as new multimodal and smaller language models enhance technologies. That said, the USPTO has issued new guidance stating that while AI alone can't be named as an inventor on patents, humans can use AI in the invention process as long as they make a significant creative contribution.  

🔮 Expert Insights from Packt Community 

LLMs Under the Hood – Building Models for Your Unique Use Cases [Video] - By Maxime Labonne, Denis Rothman, Abi Aryan 

This video course is an invaluable resource for AI developers looking to master the art of building enterprise-grade Large Language Models (LLMs). Here's why it's a must-watch: 

 Key Takeaways: 

1. Expert-Led Guidance: Learn from industry experts Maxime Labonne, Dennis Rothman, and Abi Aryan, who bring a wealth of experience in LLM development. 

2. End-to-End Coverage: Gain comprehensive insights into the entire LLM lifecycle, from architecture to deployment. 

3. Advanced Skills Development: Acquire advanced skills to architect high-performing LLMs tailored to your specific business needs. 

4. Hands-On Learning: Engage in practical exercises that reinforce key concepts and techniques for building, refining, and deploying LLMs. 

5. Real-World Impact: Learn how to create LLMs that deliver tangible business value and solve complex organizational challenges. 

Course Highlights: 

- Making informed architecture decisions for optimal performance. 

- Selecting the right model types and configuring hyperparameters effectively. 

- Curating high-quality training data for better model outcomes. 

- Mastering pre-training, fine-tuning, and rigorous model evaluation techniques. 

- Strategies for smooth productionization, proactive monitoring, and post-deployment maintenance. 

By the end of this masterclass, you'll be equipped with the practical knowledge and skills needed to develop and deploy LLMs that drive real-world impact for your organization. 

🌟 Secret Knowledge: AI/LLM Resources

🌀 Building a Scalable Foundation Model Platform for Your Enterprise: This post outlines how enterprises can provide different teams governed access to powerful foundation models through a centralized API layer. The solution described captures model usage and costs for each team to enable chargebacks. It also allows controlling access and throttling usage on a per-team basis. Building on serverless AWS services ensures the solution scales to meet demand. Whether you need transparent access for innovation or just want to understand how teams are leveraging AI, implementing a solution like this can help unlock the potential of generative AI for your whole organization.  

🌀 Making Bridges Between AI and Business: This article discusses how businesses can develop an AI platform to integrate generative technologies like RAG and CRAG safely. It covers collecting data, querying knowledge bases, and using prompt engineering to guide AI models. The goal is to leverage AI's potential while avoiding risks through a blended strategy of retrieval and generation. This overview provides a solid foundation for aligning cutting-edge models with your organization's needs. 

🌀 Evaluating Large Language Models: A Guide to Benchmark Tests: As AI language models become more advanced, it's important we have proper ways to assess their abilities. This article outlines several benchmark tests that evaluate language models on tasks like reasoning, code generation and more. Tests like WinoGrande, Hellaswag and GLUE provide insights into models' strengths and weaknesses. The benchmarks also allow for comparisons between different models. They give us a more complete picture of a model's skills. 

🌀 Code Generation Gets Smarter with Context: Google's Codey APIs now enhance code completion and generation using Retrieval Augmented Generation, which retrieves relevant code from repositories to provide more accurate responses. This "RAG" technique allows LLMs to leverage external context. The blog post explores how RAG works and demonstrates its ability to inject appropriate code snippets. While not perfect, RAG is a useful tool to explore coding variations and adapt to custom styles when used with Codey on Vertex AI.  

Partnering with Notion 

Ever tried Notion? It's a workspace that helps you do things better and faster.

You get AI for notes and teamwork, easy drag-and-drop for content, and cool new features to help manage projects and share knowledge.

🔛 Masterclass: AI/LLM Tutorials

🌀 Building a Question Answering Bot from Scratch: This tutorial shows you how to create a basic question answering bot by processing text from Wikipedia, generating embeddings with OpenAI, and storing the data in Momento Vector Index. It covers initializing clients, loading and chunking data, generating embeddings, indexing the embeddings, and searching to return answers. The bot is enhanced by using GPT-3 to provide concise responses instead of raw text. Following these steps will give you hands-on experience constructing a QA system from the ground up. 

🌀 Creating SMS Apps with Next.js and AI Assistants: This article shows you how to build a texting app that uses Next.js for the frontend and backend. OpenAI's GPT-3 is utilized to generate meeting invite messages. Twilio handles sending the texts. React components collect invite details. API routes fetch GPT-3 responses and send data to Twilio. It's a clever way to enhance workflows with AI. 

🌀 Harness the Power of LLMs Without GPUs: Google's new localllm tool allows developers to run large language models locally using just a CPU, eliminating the need for expensive GPUs. With localllm and Cloud Workstations, you can build AI-powered apps right in your browser-based development environment. Quantized models optimize performance on CPUs while localllm handles downloading and running the models. The post provides instructions for setting up a Cloud Workstation with localllm pre-installed to get started with this new way to develop with LLMs. 

🌀 Making the Switch to Open-Source Models: Hugging Face's new Messages API allows developers to easily transition chatbots and conversational models from OpenAI to open-source options like Mixtral. The API maintains compatibility with OpenAI libraries so your code doesn't need updating. You can also deploy these models to Hugging Face's Inference Endpoints and use them with frameworks like LangChain and LlamaIndex. This unlocks greater control, lower costs and more transparency compared to closed-source options.

🚀 HackHub: Trending AI Tools

🌀 arplaboratory/learning-to-fly: Train end-to-end quadrotor control policies using deep reinforcement learning on a laptop in seconds 

🌀 time-series-foundation-models/lag-llama: Open-source foundation model for probabilistic time series forecasting to perform zero-shot predictions on new time series data and eventually fine-tune for their specific forecasting needs 

🌀 noahfarr/rlx: Implements reinforcement learning algorithms using Apple's MLX framework, making the code optimized to run on M-series chips 

🌀 uclaml/SPIN: Implement Self-Play Fine-Tuning to enhance language models through self-supervised learning 

🌀 phidatahq/phidata: Open-source toolkit for building AI assistants using function calling 

Affiliate Disclosure: This newsletter contains affiliate links. If you buy through them, we may earn a small commission at no extra cost to you. This supports our work and helps us keep providing useful content. We only recommend products and services we think will benefit our readers. Thanks for your support!