1 comment

[ 5.0 ms ] story [ 11.9 ms ] thread
After explaining how Large Language Models work (like GPT), in this blog post I explain how ChatGPT works.

the content covered: - Learn how ChatGPT mastered the subtle dynamics of dialogue, from guiding frustrated users to explaining complex topics with clarity. - How Reinforcement Learning from Human Feedback (RLHF) turned ChatGPT into a thoughtful, context-aware assistant. - How "Constitutional AI" helps ChatGPT handle sensitive topics responsibly and ethically. - The Memory: Understand the mechanisms behind ChatGPT’s advanced context management, including dynamic attention and semantic linking. * See how ChatGPT generates high-quality answers by juggling goals like relevance, safety, and engagement. - Dive into the intriguing world of “jailbreaking” and what it reveals about AI safety.