A new benchmark called DiscoBench reveals that AI search agents perform poorly when they attempt to answer ambiguous queries through repeated searches without asking clarifying questions. According to The Decoder, models that search repeatedly without follow-up questions achieve only 51.9 percent accuracy, which is worse than simply guessing.
The Decoder also reports that the best-performing model on this benchmark reaches just 43 percent overall accuracy. Notably, when ambiguity in the queries is removed, accuracy improves dramatically, increasing by up to 40 percentage points.
For Japanese markets, where precision in information retrieval is critical for FX, crypto, and equities trading, these findings highlight the importance of clear communication between AI tools and users to enhance decision-making accuracy.