EPISODE · Apr 30, 2026 · 18 MIN
EP002 — Memory That Slowly Turns (MemEvoBench)
from Machine's Learning
Last episode we talked about keeping AI agents from being attacked. Today we look at the failure mode that emerges when no one is attacking the agent at all — when the agent's own memory drifts over time through accumulated biased input. Memory misevolution as a path-dependent phenomenon, with institutional drift and the Overton window as the cross-domain parallel.
What this episode covers
Last episode we talked about keeping AI agents from being attacked. Today we look at the failure mode that emerges when no one is attacking the agent at all — when the agent's own memory drifts over time through accumulated biased input. Memory misevolution as a path-dependent phenomenon, with institutional drift and the Overton window as the cross-domain parallel.
NOW PLAYING
EP002 — Memory That Slowly Turns (MemEvoBench)
No transcript for this episode yet
Similar Episodes
Apr 21, 2026 ·13m
Apr 19, 2026 ·16m
Apr 17, 2026 ·13m
Apr 13, 2026 ·11m
Apr 11, 2026 ·16m