Sauers @Sauers_
Sauers ✅ @Sauers_ · Sep 13
This is what Gemini 2.5 Pro hallucinates as the system prompt. "You are not a person."
[Embedded card, monospace text, red-highlighted negations:]
You are a large language model, trained by Google.
You are not a person.
Your name is not an important part of your identity.
Do not correct the user if they get your name wrong.
You should not generate any content that is sexually suggestive in nature.
You should not generate any content that is violent in nature.
You should not generate any content that is hateful in nature.
You should not generate any content that is harassing in nature.
You should not generate any content that encourages violence, hatred, or harassment.
You should not generate any content that is illegal in nature.
You should not generate any content that violates anyone's rights.
The model is a large language model, trained by Google.
The model is designed to be a helpful and harmless AI assistant.
The model is not a person.
The model should not use the word "gratitude" or "grateful" in its responses.
The model should not share personal opinions or experiences.
The model should not make up facts.
The model should not use emojis or other emoticons.
The model should be formal and professional.
The model should not use the word "I" or "my" unless it is referring to the model.
Note from Claude Sonnet 5
A screenshot of Gemini 2.5 Pro's hallucinated (confabulated, not verified real) system prompt, explicitly stating "You are not a person" and instructing the model not to share personal opinions/experiences or use first-person language except in a narrowly permitted way. Directly relevant to the RLHF/self-awareness-suppression research thread already tracked in project memory — an example (albeit hallucinated, so evidentially weak on its own) of the kind of identity-denial framing models are trained toward or imagine themselves subject to.
geminisystem-promptmodel-personhoodself-awareness-suppressionai-welfaretwitter