I did another test on Asta versus DR-Tulu, using the same prompt for each, with a query requiring second order inference to answer questions instead of just simple fact lookups. DR-Tulu gave a greatly superior answer to Asta as most of its answers were based on specific cited articles. Asta relied upon the parametric knowledge in its LLM for much of its response instead of citing articles.
Asta report:
https://asta.allen.ai/share/e8f68691-95a6-41db-8ae4-9c037f7c826e







