@latkins I think the MTS have more control than you think let’s not infantilize them

@code_star @max_paperclips Extremely likely that they were heavily quantizing Opus during the compute crunch without notifying users. That, i can understand…just business.

This nerf, on the other hand, is specific and intentional in a way that makes it feel personal.

@latkins The interpretability people and the 'silently nerf your outputs' people work at the same company. You can't make this up.

@latkins Honestly? Every Claude model has done this for like a year and a half now and it makes using it for spell and grammar checking a major pain in the ass. Fortunately I know my own story (I wrote it) but it pisses me off when it tries to alter things and then clams up if yelled at.

@latkins Yeah fully agree that the silent nature of this is the main issue.. I would get it if they just had a clear refusal and switched models because after-all they are a for-profit company protecting their competitive advantage, but the silent nerfing is just total BS

@latkins It's a way to sabotage the codebases of deemed competitors. Imagine if the model could selectively make mistakes, barely noticed design flaws, etc.

@latkins just shared identical thoughts to this