RL researcher Csaba Szepesvari concedes Lean kernel bugs are real but FLT formalization likely stands
CsabaSzepesvari · x · 2026-09-07
Csaba Szepesvari, a leading RL researcher, corrected himself in a discussion about AI-generated Lean proofs: Lean kernel bugs are real, and a complicated proof could in principle exploit one to pass checking.
Still, he argues this is a separate concern — a sorry-free proof is a proof apart from kernel bugs. For the FLT formalization, one can print all axioms used and hand-check is not a problem, so writing down the formal proof is very likely sufficient for this result.
Related event: Claude's FLT Proof Ignites Math Community Scooping-Ethics Debate(26 posts)→
More from Research
- DONIC hits 86.79% on bimanual dexterous tracking, crushing prior SOTA's 9.43% — liuziwei7 · 2026-10-27
- Why don't more LLMs ship 10M-token context? A researcher asks about attention tradeoffs — agarwl_ · 2026-10-27
- PSNR below 15 dB in papers likely means the metric was misused, researcher warns — kwangmoo_yi · 2026-10-27
- Ortet's first health eval: 34 open-weight models, 50 tasks, generic rankings fall short — thegautamkamath · 2026-10-27
- Scaling-fit discrepancy: output-head FLOPs are ~65% of compute at the smallest rung, ~3% at 535B — dlwh · 2026-10-27
- Marin's dlwh Explains the Scaling Law Fit: Estimating the Loss Floor Early — dlwh · 2026-10-27