Debate: Sparks versus embers
summary
Motion: whether current LLM scaling plus realistic post-training can produce new proof techniques to settle problems like P≠NP. Panelists present concrete evidence: benchmarks (MMLU from ~50% to ~86% to ~92–95%), a medical study where GPT‑4 ≈90% vs doctors 75%, and AIME math moving from GPT‑4 ≈0% to o1 ≈50% now trending higher. Methods discussed include post-training (RLHF-style extraction), o1 experiments, synthetic curricula (Phi), chain‑of‑thought, and formal checking with proof verifiers like Lean; objections focus on frequency bias in pretraining and hallucination risk over long proofs.
AI-written summary · about
chapters (7)
- 0:00Introduction and setup
- 5:44Proposition opening
- 14:01Opposition opening
- 22:10Rebuttal and discussion
- 33:51Closing statements
- 44:05Audience poll and discussion
- 58:04Final arguments
related talks
25mWhy LLMs Still Fail: Reasoning, Robustness, and Safety
19mLLMs reasoning and agentic capabilities over time
41mEnhancing LLMs with Graph Technology
01h 18mThe Enigma of LLMs: on Creativity, Compositionality, Pluralism, and Paradoxes
54mKeynote: Can you trust your (large language) model?
01h 05mCan you trust your (large language) model?
46mPlaying the long game
17mPsychology of AI