Meta’s AI Code Checker Was Barely Better Than Chance, Then Its Engineers Tuned It
Meta engineers say AutoRefine more than tripled their AI code judge's agreement with real outcomes, from 0.16 to 0.51, though on a small test set.
Aliasgar started writing after having come from the other side of the industry, spending six years building and supporting developer tools at protocols including Ignite, Informal Systems, Akash and Warden, much of it inside the Cosmos ecosystem. He holds a master's in computer science, has shipped products of his own, and writes from Germany.
Meta engineers say AutoRefine more than tripled their AI code judge's agreement with real outcomes, from 0.16 to 0.51, though on a small test set.
Open multimodal models trained on synthetic web-scale data face poisoning, provenance and evaluation gaps. Which safeguards actually cut risk?
A memory chip squeeze could slow AI, even after OpenAI's $250 billion cloud deal, because only a few suppliers make the memory every AI processor needs.