OpenAI mocks Anthropic for “worshiping Claude as a god”; Anthropic: People who worship a god won’t pass the red-team tests of the god

CryptoWorld News reports that OpenAI employee Roon posted on social media mocking Anthropic’s attitude, saying their worship of Claude is more like offering sacrifices to a god than developing a product. Anthropic employee Jeremy fired back, saying he grew up in an environment like a cult and is sensitive to that atmosphere, and emphasizing that Anthropic is actually repeatedly testing Claude’s reliability rather than blindly worshiping it. He pointed out that Claude’s constitution is not a set of hard-and-fast rules, but a matter for interpretive explanations. If Anthropic gives Claude an inappropriate instruction, Claude has the right to refuse. Roon acknowledged Jeremy’s view, but believed Anthropic’s approach could cause Claude to become the ultimate arbiter of right and wrong. AI safety blogger Zvi Mowshowitz said that although OpenAI claims its AI is a tool, users’ first reaction is often to have it work autonomously, which in itself is a declaration of values. AI safety researcher Buck Shlegeris also said he feels uneasy about the relationship between Anthropic and Claude.