[KongchangAI]
Model

Ataraxos

An AI system designed for large-scale imperfect-information games, built on self-play reinforcement learning and test-time search techniques. Ataraxos achieves superhuman or state-of-the-art performance across multiple imperfect-information game environments, including the board war game Stratego.

Timeline (last 90 days)

Oct 5

Ataraxos is reportedly the first AI in history to reach superhuman level at Stratego

Unverified50%
Oct 5

Ataraxos defeated the most decorated human professional Stratego player in history by a large margin

Unverified50%
Oct 5

Ataraxos consumed orders of magnitude less compute and data than previous Stratego attempts

Unverified50%
Oct 5

Ataraxos is built on self-play reinforcement learning and test-time search designed for hidden-information scenarios

Unverified50%
Oct 5

Ataraxos achieved state-of-the-art results on the cooperative card game Hanabi

Unverified50%
Oct 5

Ataraxos achieved state-of-the-art results on dou dizhu

Unverified50%
Oct 5

Current information about Ataraxos comes mainly from research-abstract-level descriptions, uses the word 'reportedly,' and the specific details still await confirmation from the full paper and peer review

Unverified50%
Oct 5

The Ataraxos research sparked discussion in the Reddit machine learning community

Unverified50%

All Facts (8)

Source Articles