Early GLM video-input experiments: screen recordings for UI bugs, voice still weak

DevDminGod · x · 2026-09-16

A developer shares early usage ideas for GLM's video input: recording the screen and having the model spot UI bugs works well, but voice recognition is weak—converting spoken notes to text is a better route, since the model mainly relies on the visual track.

Original post →

More from coding & agent

coding & agent channel →