“Open weights are nowhere near a sufficient solution”: Dario Amodei fires back on AI power

Anthropic CEO Dario Amodei argues that open weights, while helpful in some ways, are not enough to significantly reduce power concentration in the AI industry. He states that open weights merely shift the concentration to those with the most compute and chips, rather than providing true independence. According to Amodei, regulation is not simply a route to regulatory capture, as some Silicon Valley figures suggest, but rather can provide objective standards to constrain powerful companies.

He believes that regulatory requirements, when based on measurable standards, can help decentralize power. Amodei contends that open weights can aid in model adaptation and data privacy, but as models become larger and more complex, it becomes increasingly difficult for smaller teams to both afford the required compute and maintain control over it.

This issue is evident in recent agreements among five European companies to purchase non-existent AI compute. Anthropic, one of the few major AI labs that did not sign a July 24 letter supporting open weights, maintains its stance that open models without dangerous capabilities are a public good, but ceases to be a public good once the model can facilitate significant attacks.

Amodei proposes mandatory safety testing for models that reach this level, regardless of their release method. Anthropic also backs a call for powerful AI labs to slow down development while smaller competitors gain ground.