Papers2

This paper says we should test AI the way real life works: by letting it ask questions, gather clues, and make smart moves step by step under a limited budget.

GISA is a new test (benchmark) that checks how well AI assistants can search the web like real people do.