Could "anti-distillation" legislation backfire? Allen AI researcher says it will severely hit US open-source startups.

robot
Abstract generation in progress
ME News, April 24 (UTC+8), according to Beating monitoring, Nathan Lambert, head of the post-training team at the Allen Institute for AI (AI2) and an authority in RLHF (Reinforcement Learning from Human Feedback), warned in a post that the current push by US politicians to accelerate "anti-distillation" legislation could severely harm the US's own open-source AI startups and academic research. He acknowledged that frontier model companies like OpenAI and Anthropic are major strategic assets of the US, and that distillation indeed weakens their competitive position, but believes that if legislators act before fully understanding the actual impact of distillation, the consequences could be more severe than the problem itself. Lambert pointed out several specific risks: US startups like Cursor rely on Chinese open-source models to maintain independence from closed-source vendors; US academia has "a large amount of research built on Chinese models"; the ban would set open-source model capabilities back 6 to 12 months and concentrate power in closed-source labs. He also questioned the evidence regarding how distillation data is obtained, noting that the source is the very closed-source companies pushing for this policy. His core argument is: The US has the world's largest inference market, and cheaper open-source models form a healthy counterbalance to frontier closed-source models, driving investment and innovation. If the US now cuts ties with the global open-source community, it may lose its voice in the open-source model space. (Source: BlockBeats)
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pinned