Yann LeCun Brings Up Old Grudges to Mock Dalio: Seven Years Ago, He Even Thought GPT-2 Was Dangerous

Beating AI News Flash: Anthropic CEO Dario Amodei recently called for slowing down frontier AI, to which Turing Award winner Yann LeCun dug up old grievances from GPT-2 in 2019 to mock him. LeCun said that when Dario was still at OpenAI, he believed GPT-2 was so dangerous that it should not be open-sourced. "I mocked it back then, and everyone should continue to mock it now."

In 2019, OpenAI indeed, out of concern for misuse, did not initially release the full 1.5 billion parameter version of GPT-2, only opening smaller models first. Dario was the head of research at OpenAI at the time. Nine months later, OpenAI still released the full version entirely, stating that it had not seen strong evidence of actual misuse.