Self-improving research agents built a Hearthstone AI that climbed from 6,000 to 15,000 rating and beat the server's top ten on every statistic
AspireAI community · arXiv
AutoTavern is a self-improving research system where specialised LLM research agents share a typed research memory, every claim passes verification gates, and live games close the loop - so the agents improve the simulator, the opponents, the policy, and from the same feedback their own research process, while a human sets goals and rules on ambiguous mechanics. In 20 days it built TavernGym, the first faithful full-content reproduction of the live game, and 23 versions of TavernAgent trained with PPO against opponents calibrated to real games.
TavernAgent plays complete games on the live ladder by itself: its rating rose from 6,000 to about 15,000, and its last 20 records average a placement of 1.60 with 80% first places, better than the server's ten highest-rated players on every statistic (2.34, 59%). On TavernBench, 960 tasks in 40 scenarios, it answers 91% of play tasks against at most 63% for frontier LLMs and decides in 0.03 s where they need up to 26 s. The authors' conclusion is blunt: LLM agents build far better than they play.
The honest limit: this is one game with the agents' own reproduction of it as the simulator, so the leaderboard result depends on how faithfully that reproduction holds.
https://arxiv.org/abs/2610.33845
Read the story and explore more from the AspireAI community.
Two sparse-autoencoder analyses have resisted scaling: matching semantically similar features across layers, and compressing large feature circuits into interpretable supernodes. This paper argues both are the same underlying problem, estimating semantic distances between SAE features that sit on different activation manifolds. Its representation treats each feature not as a single decoder vector but as an activation-weighted distribution over the hidden states expressing it, projects those distributions into a shared reference space, and compares them with Wasserstein distance. The representation is proved invariant to activation rescaling and stable under perturbation, and recovers true matches under finite-sample margin conditions. Empirically it outperformed decoder-vector and LLM-based baselines and captured subtle functional distinctions between related features, and it compressed large feature circuits into interpretable supernodes automatically. The limitation is that this is interpretability tooling: the paper demonstrates better feature matching and circuit compression, not that downstream interpretability claims become more accurate.