Web search narrowed the gap between AI models in current-events tests, Braintrust says
Braintrust reports smaller gains from search for older events. It also found that a fifth search query was associated with lower performance.
TLDR
Braintrust says it evaluated 1,329 current-events questions across four models and 14 conditions, comparing You.com, provider-built-in search and no search. It reports that search reduced the gap between models from 47.9 points to 5.6. Gains from retrieving information declined with event age, from about 45 points for recent events to about 24 for the oldest. Braintrust also reports that runs with five or more searches scored 19–49%, with a fifth query associated with lower performance.
Combined views
99
1 Source, first seen 19d ago