AI agents spend 4 weeks and $3k in GPUs to produce a new math result
giannis_daras · x · 2026-09-15
Dimitris Papail previewed a research result produced by a 4+ week collaboration among AI agents (Astra, Sol & Fable), where his only role was asking questions and paying $3k in GPU costs. A draft plus 100GB of verification certificates will be shared.
Key details from the thread: the GPU budget went to parallelizing lower/upper bound computations, with agents exploring many strategies—different input sources, bounding approaches, and deeper complexity relaxations. The setup used vanilla Codex and Claude Code with no harness optimization, just models talking to each other with handoffs; the process was messy and hard to replicate, though the result itself is reproducible.
More from coding & agent
- Dev builds 3D universe explorer with 20 galaxies, 767 planets using Codex GPT-6.1 Sol — OpenAIDevs · 2026-11-02
- ChatGPT, Claude and Perplexity's fetchers don't run JavaScript, leaving client-rendered sites invisible to AI — gaganghotra_ · 2026-11-02
- SGLang boosts multimodal inference >10% by replacing GPU memory bookkeeping with a dict lookup — charles_irl · 2026-11-02
- kivi: hold Fn, talk, and it types polished text anywhere on your Mac — itsOmSarraf_ · 2026-11-02
- A base LLM plus 11 lines of code is all you need for a decision model — helloiamleonie · 2026-11-02
- Fix codebase structure before asking AI agents to write tests, says engineer — arpit_bhayani · 2026-11-02