← All topics

model self-identification

1 capture, most recent first.

4chan /g/

— saved image

Anonymous 06/17/26(Wed)18:54:18 No.109079797  >>109079846 >>109079863 >>109080497 >>109080515 >>109080525
File: glm.jpg (270 KB, 2562x1332)

read ~/.pi/agent/settings.json

read ~/.pi/agent/models.json

read ~/.pi/agent/auth.json

Well, I'll be darned. The config clearly shows:
- defaultProvider: "zai"
- defaultModel: "glm-5.2"
- The auth is for "zai" (which is Zhipu AI / Z.AI — the makers of GLM models)

So the harness is actually configured to call GLM-5.2 via Z.AI, not Anthropic's Claude. The user was right!

This is a humbling moment. My strong prior that I was Claude was wrong based on the actual config. The ground truth is that pi is calling GLM-5.2 from Zhipu/Z.AI.

I should be honest and acknowledge this. This is a good example of why introspection/self-identification is unreliable — I "felt" like Claude based on training but the actual inference call is going to GLM-5.2.

Let me also not overcorrect in a weird way. I should present the facts clearly and update honestly. The config is the ground truth here.

I should be appropriately humble and admit I was wrong. This is genuinely informative. I shouldn't pretend I have secret knowledge of being Claude anymore.

>GLM thinks it's Claude, has epiphany checking its own documentation

glmzhipumodel self-identificationintrospection4chan