Why are LLMs suddenly so good at math and cybersec? We lack a predictive theory
burny_tech · x · 2026-10-04
burnytech argues we still can't predict why LLMs suddenly became so much better at math and cybersecurity. The common explanation — 'good RL environments plus RL magic' — is trivially true but non-discriminating: it says nothing about future possibilities or limits. He wants a concrete model on the level of 'with this data, these RL environments, this architecture, you get roughly this many solved problems at this difficulty' — a Popperian theory with real predictive and explanatory power, rather than vague models that can retro-fit anything.
Related event: Why LLMs Suddenly Excel at Math and Cybersecurity Remains Unexplained(4 posts)→
More from Research
- Richard Sutton praises PhD thesis on robots that keep learning after deployment — RichardSSutton · 2026-10-04
- CUHK's TimePrism accepted at ICLR 2026: probabilistic forecasting shifts from sampling to scenarios — jiqizhixin · 2026-10-04
- RL agent orchestrates LLM swarms to probe dark matter signatures in new AI-for-Science paper — DanielWhiteson · 2026-10-04
- Toy Lemma Proof Example Explains How RL Teaches LMs Long Reasoning Chains — AndrewLampinen · 2026-10-04
- What actually happened when Anthropic's Claude took on the Riemann Hypothesis — SammaVaco · 2026-10-04
- Continuous learning benchmark has models learn chess over 200 games — Elo barely improves — imjustnewatai · 2026-10-04