Anthropic CEO: Not Advocating for Banning Open Source, But Requiring Testing Before Model Release
Anthropic CEO Dario Amodei explicitly stated that the company does not advocate banning open-weight models, but believes that every sufficiently powerful model should undergo safety testing before release. He refused to sign an industry joint letter supporting open AI and advocated three measures: preventing powerful chips and manufacturing equipment from flowing to China, stopping large-scale knowledge distillation of US models, and requiring safety testing for both open-source and closed-source models. Amodei believes that once open weights are released, safeguards can be removed and models cannot be recalled, which may benefit attackers in fields such as biology.
Read the original (opens in a new tab)
News stream data aggregated by AI HOT