from Hacker News

Mixture of Tunable Experts-DeepSeek R1 Behavior Modification at Inference Time

by pr337h4m on 5/1/25, 10:05 AM with 1 comments

• by pr337h4m on 5/1/25, 10:08 AM

  Tweet from the authors: "It was a pleasant, albeit partially intuitive, effect that the side effects of free thinking seem to be performance increasing."

  https://x.com/tngtech/status/1917847505175236850