3 ms·
What you're describing only applies to security or biotech downgrades. A downgrade related to the model believing that you're doing something related to model d
by tekacs 4mo ago
What you're describing only applies to security or biotech downgrades. A downgrade related to the model believing that you're doing something related to model development is invisible and silent and internal.
- steveklabnik 4mo agoAnthropic has reversed that decision. (But that just happened so it might have been true during the article's testing.)
- tekacs 4mo agoI was just coming here to post this reply to myself! You're absolutely right! :) Honestly so glad to see the reversal.
- espeed 4mo agoWhen I reported this, Anthropic sent me an email on Tuesday saying, "You have been approved into the Cyber Verification Program", but it's still downgrading. Is this a bug? What's the point of the Cyber Verification Program if Fable 5 downgrades when you tell it to write secure code?
- steveklabnik 4mo agoI don’t think that’s relevant? The change is that it will no longer silently downgrade, and will instead be honest that it’s doing it in all cases.
- rattray 4mo agoI think that gets you access to mythos, which doesn't have the safeguards. It's configured as a separate model.
- matheusmoreira 4mo agoNot sure if it's wise to trust them again even if they say they reversed it.
- wren6991 4mo agoThey've publicly apologised for the invisible PEFT that deliberately makes the model dumb on some tasks. Whether they still do it, or will once again do it in future in more subtle ways, is something we can't verify. Personally I think they have proven themselves to be the stewards of AI in the same way Exxon Mobil are the stewards of petroleum.