Concepttemporal generation gap
Temporal Generation Gap
The phenomenon in omni-modal models whereby the internal attention localizes the target moment with significantly higher precision than the timestamps the model actually generates, reflecting a misalignment between internal representations and output generation.
Timeline (last 90 days)
Oct 20
Omni-modal large models perform poorly on temporal grounding tasks, exhibiting a phenomenon in which the precision of internal attention localization is significantly higher than the precision of output timestamps, which the authors call the "temporal generation gap."
Unverified50%