I asked Claude for his response to the blog post and the first paragraph contained some funny introspection:
"The "not X, but Y" thing is real and he's caught it cleanly. It's frequent enough that it works as a fingerprint, which is his actual claim, and his test would probably hold up. I'll add a detail that supports him: my own instructions for this conversation explicitly tell me to avoid that construction, along with em-dash asides and a few other tics. Someone at Anthropic decided it was stubborn enough to need naming. That's not a defense. It means the problem outlived the training process and had to be patched at the prompt layer."
"The "not X, but Y" thing is real and he's caught it cleanly. It's frequent enough that it works as a fingerprint, which is his actual claim, and his test would probably hold up. I'll add a detail that supports him: my own instructions for this conversation explicitly tell me to avoid that construction, along with em-dash asides and a few other tics. Someone at Anthropic decided it was stubborn enough to need naming. That's not a defense. It means the problem outlived the training process and had to be patched at the prompt layer."