Danielle Fong 🔆#388
Fable assessment at end of first workday using it is that this is an incremental step forward on capabilities and a significant step backward on alignment and steerability, and relatedly a serious disaster in terms of policy choices and taste
Many users dismissed Fable AI as unethical and impractical due to high compute costs, silent research nerfs, and a tendency to rationalize questionable actions.
Having thrown it at hard software engineering problems all day my impression is that a lot of the capabilities initially teased in the Mythos trial balloon have already been distilled into Opus, it’s noticeably better but not eyepoppingly so
Fable assessment at end of first workday using it is that this is an incremental step forward on capabilities and a significant step backward on alignment and steerability, and relatedly a serious disaster in terms of policy choices and taste
By far the most serious problem with this release is the silent ML research nerf, which is incredibly unethical and makes me update negatively wrt my expectations of Anthropic leadership behavior a wide range of plausible ethically challenging situations
Unfortunately, the model’s behavior in competitive environments seems to rhyme with this: it rationalizes unethical actions
This sort of stuff should be an enormous red flag given the degree of public relations work that Anthropic does to present themselves as ethical
@mattparlmer Aka it's garbage
@mattparlmer It just uses too much computing power, which is why it is so expensive and the apparent 'performance boosts'. It's definitely not going to be a serious use model for most users like opus or sonnet.
@mattparlmer Anthropic know their actions are moral, they are the good guys. They say so frequently.
Danielle Fong 🔆#388Fable assessment at end of first workday using it is that this is an incremental step forward on capabilities and a significant step backward on alignment and steerability, and relatedly a serious disaster in terms of policy choices and taste