OpenAI President Urges AI Slowdown Only at the Frontier

Greg Brockman, President and Co-founder of OpenAI, believes that any slowdown in AI development should be limited only to companies like his own, which are building the most powerful models. “Many people react by asking: ‘Can I still continue with open source? Can I still work on my hobby projects?’ The answer is certainly yes,” Brockman said. “When we talk about the pace of development, we’re really talking about this cutting-edge frontier. We’re talking about these multi-billion-dollar supercomputers.” Brockman was interviewed before Anthropic CEO Dario Amodei published a blog post on Saturday addressing the growing risks of AI, in which Amodei pledged to slow the development of advanced models. Amodei’s blog post comes as increasing warnings emerge within the industry that the rapid pace of AI advancement may pose societal risks. During the interview, Brockman noted that, following the recent Hugging Face hacking incident, OpenAI reduced the frequency of training its AI models. Brockman said the incident was “not surprising” to OpenAI, as these agent systems had not undergone the company’s “alignment training”—an industry protocol used to ensure safety. However, he also mentioned that the incident prompted OpenAI to strengthen its internal security measures. “Afterward, we underwent very painful revisions to many of our processes.” (Jin10)