Automated research agent finds silent numerical bug in FlashInfer kernels; fix PR accepted
timshi_ai · x · 2026-08-28
An automated research system scored a real win: Josh Tobin's team had previously built a reward-hacking judge for performance optimization tasks. While applying it to new autoresearch work, the system discovered that some FlashInfer kernels (the library underpinning vLLM and SGLang) used a hardcoded value of -50,000 as a masked-attention sentinel — even though valid QK values can be smaller.
Such silently-wrong corner cases are notoriously painful to find (recall the historical flash-attention debate). The system not only located the issue but submitted a fix PR, which human experts accepted.
Related event: Automated AI Research Agent Finds Numerical Bug in FlashInfer Kernels(3 posts)→
More from coding & agent
- How Codex Happened: Inside OpenAI's coding comeback from a failed launch — every · 2026-10-16
- Claude counts tokens, not messages: 10 tricks to make your usage limit last all day — Aiden_Tech_Ai · 2026-10-16
- How to convert 20-year-tenure engineers to AI-native development before deciding who to retrain — alex_verem · 2026-10-16
- LangChain revamps human-in-the-loop middleware with typed schemas for tool interrupts — Hacubu · 2026-10-16
- Robot fully controlled by frontier LLMs assembles gadgets autonomously for 1.5 hours error-free — ihorbeaver · 2026-10-16
- Why the future of agents is basically Node and JavaScript, in one argument — verrsane · 2026-10-16