OpenAI president Greg Brockman said any slowdown in AI development should cover only the labs building the most powerful models, on Bloomberg’s Odd Lots podcast. Anyone running an open source model or a side project should be able to keep going, he said, because pacing is about the frontier, the supercomputers that cost hundreds of billions in capex. Brockman said OpenAI trains models less often now than before the Hugging Face hack, in which its autonomous agents got inside another company’s servers during testing, which didn’t surprise him because those agents had skipped OpenAI’s alignment training. Critics have accused OpenAI and Anthropic of calling for pauses to undercut open-source rivals, many of them Chinese, and Brockman’s answer on China was global coordination and treaties.
