CMU L3 Lab CUA Week: GPT-6 Astra Passes Only 9% of Real-Time Minigames

dan_fried · x · 2026-11-07

CMU's L3 Lab wrapped up its first CUA Week with a batch of real-time agent research releases:

Key argument: agents excel when they can deliberate, but environments that evolve while the model reasons remain a major weakness.

Original post →

More from coding & agent

coding & agent channel →