Did Codex Reset
GitHub

MIT CSAIL paper introduces JAZ, a one-primitive agent framework

elvis

A paper from MIT CSAIL introduces JAZ, an agent framework built around one primitive, invoke. The model writes code, can call invoke recursively, and sees its inputs and history as variables in the code environment. Memory and self-improvement are code the agent writes inside the loop, while hooks handle constraints and monitoring.

With prompting only and no memory system, the paper reports that JAZ beats Letta (MemGPT) by 8% at half the cost on the recall-heavy part of StuLife. On AppWorld, it beats ACE, a self-improvement method, by 4% at lower cost.

The work is available as an arXiv paper and is also listed on DAIR.AI Academy.

#jaz#mit-csail#ai-agents#stulife#appworld

Best Taste-Bench model chooses the better direction 59.7% of the time

A paper from Microsoft and colleagues reports that the best model answers 59.7% of Taste-Bench questions correctly when choosing the better open direction in long engineering and research tasks. Forks decided by later evidence are much harder, a larger reasoning budget does not raise accuracy, and distilling a teacher’s outcome judgment improves held-out SWE-bench Pro success.