Eric Bigelow warns open-source AI restrictions could hinder interpretability research

What was said

"I think there's like there have been some loose discussions around like placing restrictions on open source models bar none, you know, regardless of of where they're coming from. And I think that would be the bigger that would be a real problem."

Eric Bigelow says proposals to restrict access to open-source AI models, regardless of where they are made, would create major obstacles for interpretability research. He argues that safety-relevant work on higher-capability behaviours, including reward hacking and other misalignment failures, depends on access to large open models.