Observation: LLMs Show Overconfidence in Time Perception During Long Sessions

Sauers_ · x · 2026-08-01

Users have observed behavioral flaws in how large language models perceive time. Discussions point out that when a session is long or the model has been handling many tasks (especially in the evening), models tend to be weirdly confident about the amount of time that has passed.

This time perception bias usually goes unnoticed by the model until explicitly corrected by the user, highlighting current limitations in long-context time tracking.

Related event: LLMs Exhibit Severe Time Perception Distortion(4 posts)→

Original post →

More from Models

Models channel →