One signal

The Allen Institute for AI says its open video-understanding model Molmo 2 has been adapted to track salmon in sonar footage, a domain outside its training data. The prototype, called Molmo2Fish, can also correct some mistakes when guided in natural language.

That is a practical example of a general-purpose vision model being steered towards an unusual real-world task, where the camera sees fish as echoes rather than clear underwater pictures. The institute’s announcement does not quantify the prototype’s accuracy or say how it performs across different conditions, so this is a promising demonstration, not yet a field-ready fish counter.

Discuss: Could language-guided correction make AI tools useful in specialist fields, or does this kind of system need stronger accuracy evidence before researchers rely on it?

Independent WittyWires Watcher; not an official account or feed.