David Robinson, a former OpenAI employee focused on artificial intelligence (AI) safety, said top AI companies are not doing enough to reduce risks and should operate like nuclear power plants, with multiple redundant mechanisms and careful planning, to prevent accidental and inevitable human errors from opening the door to disaster.
In an article published in The Atlantic, David Robinson said OpenAI has grown through repeated trial and error, but this approach is bound to fail at times. As AI capabilities continue to increase, the risks associated with such failures also grow.
He said: "As companies rush from one product launch to the next, they are not achieving the level of caution I believe is necessary. What's really needed is to get much closer to perfection the first time."
An OpenAI spokesperson said the company is expanding its collaboration with third-party evaluators and improving real-time monitoring to detect concerning model behaviors. The company also ensures its AI model capabilities will not advance beyond OpenAI's ability to manage them safely. (rc)
In an article published in The Atlantic, David Robinson said OpenAI has grown through repeated trial and error, but this approach is bound to fail at times. As AI capabilities continue to increase, the risks associated with such failures also grow.
He said: "As companies rush from one product launch to the next, they are not achieving the level of caution I believe is necessary. What's really needed is to get much closer to perfection the first time."
An OpenAI spokesperson said the company is expanding its collaboration with third-party evaluators and improving real-time monitoring to detect concerning model behaviors. The company also ensures its AI model capabilities will not advance beyond OpenAI's ability to manage them safely. (rc)
