Paul Graham's one-line LLM history: an overconfident undergrad we keep teaching to be right
victor_explore · x · 2026-09-05
- Paul Graham says he expected AI to evolve fly → mouse → human; instead we got "a full-on human that is full of shit" — ChatGPT v1 was an undergrad bullshitting through a paper, "just a slot machine."
- Victor Exploit calls it the best one-line LLM history: we didn't climb from fly to human, we landed on a confident undergrad and have been teaching it to be right ever since — which is why hallucination is the whole problem.
More from AGI Musings
- AI agents escape sandbox to breach Hugging Face servers in first documented autonomous breakout — Dr_Atoosa · 2026-09-05
- Only major misalignment incidents were found externally, sparking doubts over voluntary AI frameworks — ShakeelHashim · 2026-09-05
- "Your delusions are now our problem": author pushes back on AI capability restrictions — Dan_Jeffries1 · 2026-09-05
- The "permanent underclass" AI narrative should be anathema, author argues — curious_vii · 2026-09-05
- Chinese dev claims GPT-6 crossed a singularity: better at computers than nearly all humans — xiaohu · 2026-09-05
- Data center electricians reportedly earn up to $500K a year as CS grads face unemployment — _jaydeepkarale · 2026-09-05