VALG: An open-source agentic system for automated ML theory research
YiMaTweets · x · 2026-08-17
VALG is an open-source agentic system designed for automated ML theory research, converting the iterative rhythm of theoretical work into a structured workflow. A key feature is joint problem solving and problem reformulation, iterating between adjusting assumptions and deriving proofs.
Tested on 5 COLT 2026 open problems, VALG produced 22 theorem candidates across 9 subproblems. Two were fully solved, and seven were partially resolved via relaxations. The team is also building a benchmark for ML theory automation, with initial results suggesting around 80% of arXiv 2026 theory papers can be handled end-to-end.
More from coding & agent
- Writing C code manually with AI autocomplete beats AI generation — tetsuoai · 2026-08-17
- LlamaIndex releases LlamaExtract agent with 94% accuracy on long docs — llama_index · 2026-08-17
- Dizko open-sources city-events MCP server with 18 tools across 32 cities — eye_tawnyah · 2026-08-17
- Dev Rant: Juggling Multiple AI Chats is More Tiring Than Coding — yihui_indie · 2026-08-17
- Open-source zxLLM predicts LLM VRAM usage & KV-cache needs with high precision — Capable_Item_5918 · 2026-08-17
- Hermes Agent fixes desktop media stream credential leak — Teknium · 2026-08-17