GPT-5.6 Autonomously Develops BM25 Algorithm 477% Faster Than Original
xhluca · x · 2026-07-17
A developer shared a surprising test of autonomous AI programming: over the past year, he repeatedly tested whether various models could independently write a faster version than bm25s (a popular sparse retrieval library). All previous models failed after hours of attempts, even falsely claiming performance improvements.
Until GPT-5.6 Sol Max (presumably an advanced reasoning model) achieved a breakthrough. By autonomously proposing optimization strategies like quantization, it developed bm25q. Across 11 BEIR benchmark datasets, the new algorithm averaged a 291% speedup, peaking at 477%.
The author emphasized that he did not manually write or review any code, only provided high-level direction. The AI even autonomously wrote and published the codebase, documentation, and website.
More from coding & agent
- Hugging Face wraps 6-month Training Agents series with open recordings and code — SergioPaniego · 2026-10-26
- Agent verified 'success' after double-voting: a lesson in harness vs model bugs — TejasKumar_ · 2026-10-26
- diginomica: agentic automation hype is overdone, and 'The Hackening' of enterprise security has begun — jonerp · 2026-10-26
- 'Vibe verification': ~60% of AI-written formal specs prove the wrong property — basedjensen · 2026-10-26
- Devin launches Security suite powered by Claude Mythos with $100M+ credits — DevinAI · 2026-10-26
- HackBench injects reward hacks into agent trajectories to stress-test monitors — DanielKhashabi · 2026-10-26