Resolution autoformalizes math to speed AI alignment research
Resolution, an AI safety startup, published work on autoformalizing mathematical statements related to statistical learning theory. The team used agents to develop plans—one high-level and one detailed—and an orchestrator agent executed the work, producing 600k lines of Lean code in about a week. The process cost roughly $135,000 in token spend and involved 100B input tokens, with 300M output…
Key points
- Resolution autoformalized math on statistical learning theory using AI agents, producing 600k lines of Lean code in a week
- Process cost ~$135k and consumed 100B input tokens, 300M output tokens, with 90% cache reads
- Team says work accelerates AI alignment research by automating theory and improving model generalization understanding
The effort highlights how AI agents can now handle tasks previously too expensive or time-intensive for human researchers. Resolution notes the work has already sped up other aspects of their research, such as automating parts of singular learning theory to better understand sampling. The startup’s approach demonstrates how AI tools are reshaping frontier research, potentially enabling rigorous automation of scientific discovery.
Model pages: Claude Opus 5.5 → · GLM 5.3 → · Kimi K3 →
Import AI 478: Math for alignment; ProgramBench; and Mirendil shows what RSI means
Import AI · 26 October 2026
Loading the full article…
This text was published by Import AI and written by Jack Clark. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Graphite finds ChatGPT still rarely mentions many brands even when using search · 1 src
- Google DeepMind’s AI model predicts Hurricane Isaias’ Florida landfall more accurately than conventional systems · 1 src
- OpenAI’s internal model solves Navier-Stokes using a swarm of 10,000 AI agents · 1 src
- Researchers release Hi-TPH dataset and Hi-TPH-PLMs for T cell receptor–pHLA binding prediction · 1 src
- Researchers release LAMB, a Latin language model trained on 74B tokens · 1 src
Comments
via GitHub Discussions