Princeton researchers develop LLM achieving 2700 Elo in chess

VERA Brief

AI-generated. Grounded in the article and its cited sources.

Princeton researchers have developed a large language model with 4 billion parameters that achieved a chess Elo rating of 2700. The model can also explain its strategic decisions, indicating potential for AI transparency.

Key facts

• A large language model with 4 billion parameters was developed by Princeton researchers.
• The model achieved a chess Elo rating of 2700.
• The model demonstrated the ability to explain its strategic decisions.
• The training methodology may be applicable to other domains.
• The model showed no signs of plateauing in performance when training was halted.