Jev CEO's 18-minute talk: the ChatGPT era is over
Diogo Almeida draws the line between assistance AI and automation AI, and explains how Jev replaces human approval with calibrated decisions.
Jev CEO: "I co-authored GPT-4, ChatGPT and RLHF - and I'm one of the few people at OpenAI who actually hates on ChatGPT." Diogo Almeida then spent 18 minutes on stage explaining why the ChatGPT era is over, why Claude Code is still part of it, and what he is building instead. Nobody has dropped this talk on Twitter yet. You are early. In this 18-minute video: 4:16 - the divide nobody names: assistance vs automation. Today's AI is incredible at pleasing the human in the loop - and useless at running without one 7:16 - "overpromising is a feature, by design." He feeds ChatGPT a fart sound effect labelled "the music I made" and it calls it an eerie, atmospheric piece. That is RLHF working exactly as intended 15:28 - the bitter lesson, inverted: the task matters more than the data, and the data matters more than compute. Then the third optimization - calibrated decisions instead of human approval. That is Jev The man who co-authored the greatest hits just called the era he created a detour. Watch it before your feed does. Eighteen minutes that will reprice everything you are building. Bookmark it - his follow-up teaser ("the original scaling laws were incorrect") lands on his Twitter @CompleteSkeptic, and you will want the context when it does.
This post quotes another post: x.com/0xmovez/status/2101007482919227841