TypeSafe Wants AI to Disappear

🔥 What happened Diogo Almeida — ex-OpenAI researcher and InstructGPT co-author — launched Jev via his startup TypeSafe AI. The model ditches chat in favor of "Reinforcement Learning for Calibrated Decisions" (RLCD), optimizing for honest probabilities instead of human ratings. 💡 Why it matters RLHF rewards sycophancy and hallucinations; RLVR only solves verifiable tasks. RLCD claims to sidestep both and embed AI into software dependencies — not as a chatbot, but as invisible infrastructure. Almeida's thesis: tasks and data beat compute. ⚡ Our take Finally someone admits chat-tuned LLMs are a hack on top of a string paradigm. Whether RLCD delivers is another question — but the direction is right.