I’ve been facing an issue where the AI chat forgets context upon refreshing or reopening the browser. I want to maintain chat history and system prompts for seamless conversation continuation. I need help applying a prompt to a previous conversation to resolve this problem. Can this be done? Thanks in advance!
OpenAI set a limit on the memory of ChatGPT, no small part of which is a deliberate safety limit. It can handle around 4,000 tokens, which is very roughly 3,000 words, which is enough to give it plenty of context, but not enough to retrain it in dangerous or malicious ways that earlier chatbots were destroyed with (see Tay (chatbot) - Wikipedia ).
The safety limits ensure that the core LLM and AI can’t be retrained by users, not even if thousands of users all worked in unison - as the chats never, ever become a core part of its learning. Instead, each user gains access to the AI with a personalized ‘working context’ which is the last 3,000 words of your entire chat history. That’s a slight oversimplification of a complex system, but good enough to work with.
2 Likes