Interpolation vs exploration: the core dispute over whether RL lets LLMs create genuinely new math
just_steve_h · x · 2026-10-09
The crux of the debate: juststeveh argues LLMs merely interpolate through existing material, citing Warren McCulloch's cybernetics argument that—like the second law of thermodynamics—computing cannot produce more information than it receives. tdietterich counters that reinforcement learning lets models explore search paths absent from their sources, like AlphaGo discovering new winning moves: "these systems are not just recombining existing material."
More from Research
- MathEconLib open-sourced: Axiom Math and top economists bring machine-checked proofs to economics in Lean 4 — soumitrashukla9 · 2026-10-23
- Ken Ono invites the community to build MathEcon on the new Lean 4 economics library — soumitrashukla9 · 2026-10-23
- IFM open-sources RL360: 8,192 concurrent sandboxes and ~1.3M rollouts for agentic RL — rupspace · 2026-10-23
- Science lifts expression of concern on 2024 AI debunking-conspiracies paper; conclusions unchanged — DG_Rand · 2026-10-23
- Embodied First Exam benchmarks coding agents on 1,090 tasks across 44 robots — ManlingLi_ · 2026-10-23
- Anima Anandkumar on neural operators: AI weather models beat supercomputers — AnimaAnandkumar · 2026-10-23