Analysis
This article provides a fascinating glimpse into the complex architecture behind the Generative AI models we use every day! It highlights the critical difference between the raw Large Language Model and the refined assistant we interact with, emphasizing the importance of alignment and safety in creating user-friendly AI.
Key Takeaways
- •The Generative AI we interact with is a highly processed version of the underlying Large Language Model.
- •Multiple layers, including instruction following and safety constraints, are added on top of the base model.
- •Understanding these layers helps users appreciate the complexities of modern AI systems.
Reference / Citation
View Original"In other words, what people usually see is not AI itself, but rather the output of an assistant that has been adjusted through multiple layers for use in human society."