Le nouveau modèle d'IA JEV enflamme le web : vrai ou démo truquée ?
Shubham Sharma evaluates Jev across real automation scenarios including batch email sorting, video frame text moderation, inbound scoring, and edge-case classification failures.
Original by Shubham SHARMAEvaluationIntermediate15 min 13 secPublished
Jev evaluates supplied options and returns probabilities rather than a conversational explanation.
Shubham’s screening workflow extracts text from video frames before asking Jev whether that text contains sensitive information.
The counting and date examples produce errors; Shubham recommends checking confidence rather than assuming every typed decision is correct.
Worth knowing
These are exploratory demonstrations, not an accuracy or privacy audit. The suggested confidence cutoff is not a validated safety threshold, and the video-text workflow uses separate text extraction rather than native Jev vision.