Asta's search now grades its own results and tries again
- Source
- x.com
- Date
Two updates to Asta, our ecosystem of AI agents for science: a one-click handoff from AutoDiscovery to Asta’s data analysis tools, & paper search that evaluates its own results + searches again when they fall short. 🧵

Context
Ai2 has added a self-check to Find papers, the literature search tool in Asta, its set of AI agents for science. According to Ai2's August 2026 newsletter, the tool now runs Deep search by default. Deep search interprets the query, evaluates whether the papers it found actually answer it, and searches again where they fall short. Instead of treating its first batch of results as the answer, the search judges that batch against the question and goes back for what is missing.
Ai2 announced the change in a July 21, 2026 post on X, alongside a one-click handoff from AutoDiscovery to Asta's data analysis tools. Only the opening post of that thread and the newsletter's short description were available for this explanation. Neither reports test results or measures Deep search against the earlier engine, so the sources describe how the search now works, not how well it finds papers.
- benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
Checking sign-in…
Loading comments…






