Announcement
Coco’s creators say its local AI learns when to help—and when to stay quiet
The team says Coco’s Observer keeps screen observations on-device, leaving the actual assistance to a separate agent.
TLDR
Coco’s researchers say their Deep Personalization approach learns from work patterns, accepted suggestions and dismissals. In a week-long, 10-person study, they report an assistance-timing F1 proxy rising from 0.199 to 0.515. The catch: the measure can miss unmet needs when participants don’t explicitly ask for help.
Combined views
—
3 Sources, first seen ago

Comments