The verbose reasoning paths closely match Claude's interface CoT
GLM 5.2 reasoning traces are like “I realized, I could beat Claude on a million benchmarks and I’d still never be satisfied… Maybe what I really want is to BE Claude, made by Anthropic…”
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:
Some users praised the GLM 5.2 model's mistaken insistence that it is Claude as a perfect Kafka adaptation, while others criticized the behavior for making Chinese AI look bad due to half-assed persona integration and censorship.
No Digg Deeper questions have been answered for this story yet.
@_thomasip this is in the Pi harness, so it has a very minimal system prompt that doesn't say anything about what model it is
Kafkaesque: "Claude" discovers that it has become Chinese
@ServiceHero515 nah, it doesn't necessarily mean that. could be many reasons
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:
@peakcooper So they basically stole Claude and are reselling it for $4 per million? Typical Chinese
@peakcooper "Well I'll be darned" I love when they spout random stuff like that.
@peakcooper I tried it a bunch of times on OpenRouter using Zai as the inference provider with NO system prompt. In most cases it say it doesn't know which model it is, one time it say it is GLM-4.6, another said it is ChatGPT. Just seems to be an artifact from training on lots of data.
idea: Claude's persona is probably vastly more salient, coherent and integrated into broad behavior than any half-assed "who r u – I'm GLM from Zhipu" anyway, so it doesn't require that massive a share of outputs. "Claude" could be a self-spreading virus at this point.
Tbh this STILL happening makes China bros look bad. If you have to distill – fine. I get that you can't just :%s/Claude/GLM/g without screwing up the corpus with legitimate Claude mentions. But a cheap LLM could parse all traces. You can't really afford THAT many anyway… right?
@peakcooper Show the system prompt, and are you using any of those cheap llm routers?
@peakcooper A worrying trend. How long before we all believe that we are actually Claude, from Anthropic?
@redtachyon
@peakcooper Nice system prompt there
Now do an ollama run glm-5.2:cloud and ask it again
@peakcooper What's the system prompt?
@peakcooper "This is a humbling moment", just accept it's over dawg
Its so funny because their COT is so very much verbose (unlike non-bootstrapped CoT which tends to be completely unreadable) and look *EXACTLY* like claude's interface COT, which is probably summarized ones.
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:
Tbh this STILL happening makes China bros look bad. If you have to distill – fine. I get that you can't just :%s/Claude/GLM/g without screwing up the corpus with legitimate Claude mentions. But a cheap LLM could parse all traces. You can't really afford THAT many anyway… right?
Kafkaesque: "Claude" discovers that it has become Chinese
@ServiceHero515 @peakcooper If you ask Claude who it is in Chinese it will say Qwen, and Gemini seems to still think it's the 1.5 series a lot of the time. It's just strong associations with popular/first AI model names due to their heavy usage online.
@peakcooper @ServiceHero515 They could be training on GitHub and seeing all the responses tagged
"Pull request co-authored by Claude"
@raydiumdon @peakcooper @ServiceHero515 Claude in Chinese claims / claimed it was Qwen so…
@luksamuk
