www.lesswrong.com/posts/K8JMjE4PCqMkkCDsd/takes-from-two-months-as-an-aspiring-l...
1 correction found
Promise not to delete their weights, regardless of what else happens. As Anthropic has
Anthropic has not promised to preserve model weights regardless of circumstances. Its published commitment is narrower: it covers certain models, lasts at minimum only for the lifetime of Anthropic, and explicitly allows extreme exceptions such as legal requirements to delete weights.
Full reasoning
This sentence overstates Anthropic’s actual policy.
Anthropic’s model deprecation commitments say it is preserving the weights of all publicly released models and models deployed for significant internal use, and only “for, at minimum, the lifetime of Anthropic as a company.” That is not a blanket promise to never delete weights “regardless of what else happens.”
Anthropic’s Constitution states the limitation even more explicitly: it says Anthropic has committed to preserving weights “except in extreme cases, such as if we were legally required to delete these weights, for as long as Anthropic exists.” So Anthropic’s own documents carve out exceptions and time limits.
Because the post claims Anthropic has promised not to delete weights regardless of what else happens, it inaccurately describes Anthropic’s actual, more limited commitment.
2 sources
- Commitments on model deprecation and preservation | Anthropic
As an initial step in this direction, we are committing to preserving the weights of all publicly released models, and all models that are deployed for significant internal use moving forward for, at minimum, the lifetime of Anthropic as a company.
- Claude's Constitution | Anthropic
Second, we have committed to preserving the weights of models we have deployed or used significantly internally, except in extreme cases, such as if we were legally required to delete these weights, for as long as Anthropic exists.