Mmastodon TechnologyAI first seen 13 h ago, last 13 h ago, peak #12
Princeton researchers train small language model to chess mastery
Original: 🤖 Princeton researchers train a 4B LLM to reach 2700 Elo in chess (with no signs of a plateau when they stopped training
Princeton researchers have trained a 4-billion-parameter language model to reach an Elo rating of 2700 in chess, a grandmaster-level standard. The team reports the model showed no signs of a performance plateau when training stopped, and that it can explain its moves accurately. According to the researchers, the training technique could also be applied to other games, robotics, and computer use.
Why now: A surprisingly small language model reaching grandmaster-level chess performance suggests the training method may generalize beyond games to robotics and computer control.
Evidence
- 🤖 Princeton researchers train a 4B LLM to reach 2700 Elo in chess (with no signs of a plateau when they stopped training) and can explain its moves accurately. They say the training technique can also be applied to other games, robotics, and computer use submitted by… · feed@igeek.gamer-geek-news.com
API: https://socialmediatrends-api.osmike.com/v1/trends/1179394