Research
AutoTavern agents build a Hearthstone Battlegrounds AI that outperforms top human players
TavernAgent's rating reportedly rose from 6,000 to about 15,000 on the live ladder, with 80% first places in its last 20 recorded games. TavernBench results show it answering 91% of play tasks against at most 63% for frontier LLMs; the code is slated for open-source release.
Every outlet that ran it 1 article
Hugging Face PapersResearchLead