Cohere Labs talk: Kernel Autoresearch uses coding agents to auto-discover valid ML kernels
Cohere_Labs · x · 2026-11-05
Cohere Labs Open Science Community announced its Nov 9-14 events, featuring CUHK-Shenzhen PhD candidate Richard Cornelius Suwandi presenting "Kernel Autoresearch (Kernaut)": coding agents write kernels as programs for automated model discovery.
- Kernel design faces a trade-off: a fixed grammar of base kernels guarantees validity but limits search; free-form LLM-generated programs often break — in tests, 22-58% of kernels passing simple numerical checks fail at different scales or dimensions
- Kernaut uses construction contracts to ensure every accepted kernel is valid, a quality-diversity archive to keep strong kernels with different behaviors, and novelty screening to push agents toward genuinely new ideas
- Discovered kernels transfer to unseen tasks (black-box optimization, enzyme kinetics) without further LLM calls, remaining interpretable programs humans can refine
- Paper: arxiv.org/abs/2610.10394; the speaker co-developed OpenEvolve, an open-source evolutionary coding agent
More from coding & agent
- Voice prompting in Antigravity + Gemini 4: rambling your creative decision tree to the model — alexanderchen · 2026-11-05
- Voice prompting in practice: one long spoken prompt drives a full dev iteration — alexanderchen · 2026-11-05
- DrawSimple for Mac ships with an MCP server so agents can draw editable mockups — JordanRunsForFun · 2026-11-05
- Dev builds a working keylogger in 15 minutes, then can't reliably detect it — MariyaSha888 · 2026-11-05
- Gradium and SurrealDB demo a fully on-device research agent with memory and streaming TTS — mattturck · 2026-11-05
- webIQ demo shows the evidence object returned to models, stressing freshness signals — VeryWellVersed · 2026-11-05