Summary

In this chapter, we introduced the concept of prompt engineering. I’d define that as everything that ekes out accuracy gains from your model without updating the weights of the model itself. Said another way, this is the art of getting more with less. We walked through few-shot learning, where you send a few examples of your desired inference results to the model, to zero-shot learning, where you hope to get a response from the model without any prior information. Needless to say, consumers tend to strongly prefer zero-shot learning. We covered a few tips and tricks for prompting text-to-image models, especially how to get good performance out of the open source Stable Diffusion. We learned about image-to-image prompting, where you can pass images to your diffusion-based models to produce a new image using an intersection. We also learned about prompting LLMs, including the implications of instruction fine-tuning, chain-of-thought prompting, summarization, and defending...