Tag
Introduces Flows, an execution and verification layer for software-building agents, arguing that agents need a separate completion signal alongside model confidence, demonstrated by a multi-module app with 59/59 checks passing.
Jane Street, previously skeptical about formal methods, is now building a team to use them, driven by AI and agentic coding that reduce costs and increase benefits for software verification.
A new paper surveying formal methods practitioners on the importance and tractability of applications to AI safety, accompanied by a broader plea for ambitious software verification.