People love defending Anthropics shortcomings…

“Mythos isn’t supposed to be that good at security, because actually Anthropic was referring more about running llms than mythos specifically”

“The opus model is worse because they have no compute because they are training mythos. The degraded performance is justified!”

“All the bugs in Claude code is just because the models are so good they are just looping and are shipping fast”

Constantly see people crawl out of the woodwork to defend a trillion dollars company overhyping every press release it gives

Actually, OpenAI made a similar claim about one of their GPT models a while ago…

Funnily enough that was while Dario Amodei was their research director.