Very good introduction on - let’s say - the gap between promises and reality of #LLM and #chatGPT.
Many call it #slop or #microslop. The scientific term is #Bullshit thanks to #HarryGFrankfurt and these guys.
CZM Rewind: The Academics That Think ChatGPT Is BS (https://pods.link/ep/i/1000721751470)
@iamlayer8 True. The model itself is stateless. The state (or context if you want) is all that the services memorizes and adds that to all requests. And that part is making those services more and more valuable.
Which goes in both ways. Usefulness for the user and better data for the provider. I agree privacy and information ownership is a major concern here.
Reading this blog post I learned how little innovation beyond the LLM itself was probably needed to get here: https://manthanguptaa.in/posts/chatgpt_memory/
When I asked ChatGPT what it remembered about me, it listed 33 facts from my name and career goals to my current fitness routine. But how does it actually store and retrieve this information? And why does it feel so seamless? After extensive experimentation, I discovered that ChatGPT’s memory system is far simpler than I expected. No vector databases. No RAG over conversation history. Instead, it uses four distinct layers: session metadata that adapts to your environment, explicit facts stored long-term, lightweight summaries of recent chats, and a sliding window of your current conversation.
Arnd Layer
Roman