Automating AI Without Fine-Tuning: 6 Techniques That Broke a 56-Year Math Record
theomitsa · x · 2026-08-02
Researchers summarized 6 techniques to automatically improve AI systems without fine-tuning weights or manual intervention. They all share a basic loop: an LLM proposes a change, an evaluator scores it, and only improvements are kept.
They differ in what they edit (text, prompt, code, or training loop) and how they judge it. Results are striking: one method broke a 56-year-old math record, and another beat RL using 35x fewer rollouts.
More from Research
- It's not gradient descent: burny_tech corrects the diagnosis to RLHF mode collapse — burny_tech · 2026-10-16
- A beginner-friendly introduction to causal inference kicks off, Part 1 — Nasereliver · 2026-10-16
- FBI method trains for Pass@K with one sample per update at 12.3% of the cost — EmtiyazKhan · 2026-10-16
- Berkeley researchers sample humanoid robot policies directly from weight space with diffusion — berkeley_ai · 2026-10-16
- OneWM: unified causal training shows world and action modeling improve each other — IAAI233 · 2026-10-16
- Baidu's Virtual Head Attention tames the KV cache, FLOPs, quality triangle in long-context LLMs — baidu · 2026-10-16