Hunt Instead of Wait: Evaluating Deep Data Research on Large Language Models

In three linesNew DDR-Bench benchmark evaluates investigatory intelligence of LLMs: autonomous ability to explore databases and extract insights without explicit queries. Frontier models show emerging agency but struggle with long-horizon exploration. Study distinguishes investigatory intelligence (setting own goals) from executional intelligence (completing assigned tasks).

Summary generated by Claude — human-verified