IT Home On October 8, according to a report by Business Insider early this morning (the 8th), David Robinson, one of the former leaders of the security team who left OpenAI last week, believes that the company has clear inconsistencies in its actions and words regarding AI security issues.
Robinson said, OpenAI repeatedly warns about the risks that increased capabilities of new models may bring, while openly discussing slowing down AI development, yet the company is still advancing at full speed. “What really confuses me is that we keep issuing these warnings, and in the end, we are still training and deploying those models that we warned about being dangerous.”
In July this year, hundreds of OpenAI employees signed a public lettercalling on the government to implement regulatory measures to slow down AI development.
Previously, Anthropic CEO Amodei called on AI companies to coordinate with each other to slow down development. In September this year, OpenAI CEO Oltman agreed with this view. “We need to control the pace of advanced AI development. These past weeks, this has been one of the main topics discussed within OpenAI.”
Robinson said that the OpenAI office is still in a state of rapid activity, and many people are almost relying on overdrawn energy to carry out their work. He described this state as follows: “It feels almost like ballet, or some kind of dance, with all departments’ work converging at the same time. The entire process is filled with adrenaline.”
OpenAI researchers do not hide the dangers that may arise from allowing AI to improve its capabilities on its own. This process of continuously enhancing AI’s abilities is called “recursive self-improvement”. In September this year, Yakub Pajotzki, OpenAI’s chief scientist, warned in a blog post discussing recursive self-improvement: “Extreme caution is needed at present.”
According to previous reports by IT Home, on the 2nd local time, a spokesperson for OpenAI confirmed that David Robinson, one of the leaders of OpenAI’s security team, has resigned.
Related Reads: