Summary of Mastering Board Games by External and Internal Planning with Language Models, By John Schultz et al.
Mastering Board Games by External and Internal Planning with Language Models
by John Schultz, Jakub Adamek, Matej Jusup, Marc Lanctot, Michael Kaisers, Sarah Perrin, Daniel Hennes, Jeremy Shar, Cannada Lewis, Anian Ruoss, Tom Zahavy, Petar Veličković, Laurel Prince, Satinder Singh, Eric Malmi, Nenad Tomašev
First submitted to arxiv on: 2 Dec 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: Computation and Language (cs.CL); Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary A novel approach combines large language models (LLMs) with search-based planning to significantly improve their performance in complex tasks such as multi-step reasoning. The authors introduce two methods: external search, which uses Monte Carlo Tree Search (MCTS) guided by the LLM without an external engine, and internal search, which generates a linearized tree of potential futures and a final choice within the LLM. Both approaches build on a pre-trained language model with domain knowledge, capturing transition and value functions across various board games (Chess, Fischer Random/Chess960, Connect Four, and Hex). The pre-training method minimizes hallucinations by accurately predicting states and legal moves. Results show that both internal and external search improve win-rates against state-of-the-art bots, even reaching Grandmaster-level performance in chess while operating on a similar move count search budget per decision as human Grandmasters. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models can play games like chess or Connect Four well, but they struggle with planning and reasoning. Scientists found a way to make these models better at planning by combining them with special kinds of searches. They tested two methods: one that uses an external search tool and another that does the searching inside the model itself. Both methods used language models trained on information about the games. The results showed that both approaches made the models much better at playing the games, even beating grandmasters in chess. |
Keywords
» Artificial intelligence » Language model
