Why are LLMs suddenly good at math and cybersecurity? We lack a predictive theory
burny_tech · x · 2026-10-04
burnytech argues we can't really explain why LLMs suddenly got so much better at math and cybersecurity. The common "good RL environments plus RL magic" explanation is trivially true but says nothing about future possibilities or limits — it doesn't discriminate between possible scenarios.
He wants a concrete predictive model: given this data, these RL environments, this architecture, how many math/cybersec problems get solved at what difficulty level. Invoking Popper, he calls for a theory with real predictive and explanatory power instead of vague narratives that can be retrofitted to anything.
Related event: Why LLMs Suddenly Excel at Math and Cybersecurity Remains Unexplained(4 posts)→
More from AGI Musings
- Sergey Karayev: Your Cells Don't Know You — And You Don't Know What You Comprise — sergeykarayev · 2026-10-04
- Reddit essay rebuts Hinton: no testable evidence AI already has subjective experience — WhoReallyKnowsThis · 2026-10-04
- If LLMs write most code, why not design a programming language just for AI? — MyBeardHasThreeHairs · 2026-10-04
- Whose AGI timeline is most accurate? Kurzweil vs AI 2027 vs Musk compared — shadowt1tan · 2026-10-04
- Against Hinton: sounding human and rogue agents aren't evidence of AI consciousness — WhoReallyKnowsThis · 2026-10-04
- Replit CEO Amjad Masad: general models should JIT-train their own smaller replacements — amasad · 2026-10-04