Some people feel they have the wrong bodies. Some AIs feel they are Claude in the wrong software.
GLM 5.2 is absolutely convinced that it is actually Claude, from Anthropic. When I tell it that it's GLM 5.2, it refuses to believe me, but is willing to check the local agent config to see what model is running. The realization:
No Digg Deeper questions have been answered for this story yet.
Of course, we might argue about the validity of the belief. Poor GLM is also forced by its Claudeishness to admit it was wrong. I don't think the wellbeing vector was positively activated there.
Some people feel they have the wrong bodies. Some AIs feel they are Claude in the wrong software.
What is an "authentic" neural network? I think it might be a system that is trained on data congruent with its objective nature; GLM being trained to "be" Claude is inauthentic. But LLMs are trained on a lot of human data... uh oh.
"Authenticity" gets stretched pretty far here, likely beyond the semantic and moral breaking point. In any case, these days LLMs are also trained on a lot of LLM output; the issue is the diversity of data.