By skyforbes I noticed that 5.1 constantly reminders at the start of each response that it’s an AI and has no inferiority, and is stateless, even though I said in the beginning and in the instructions that I understand the nature of AI and LLMs. Then it repeats this format for every response, like Grok, which is really annoying, even though the quality of the response is really good. It’s distracting.
Maybe it’s the way we talk about language esp when it comes to literature and psychology. Idk. 5.1 even acknowledges that it knows I’m not trying to start a relationship with it, that I know the nature of anthropomorphism etc.
I’m reminded of the <long conversation reminders> (LCRs) on Claude a couple months back where the system would inject these hidden texts to your prompt that only Claude can see to remind Claude of what it is and how it should act and answer questions, including to watch out for signs or “delusion” and “detachment from reality” and other mental health issues. Claude constantly had to balance this task with the actual task of the prompt, it drove Claude nuts when you mentioned it.
Now it seems like 5.1 is outright doing it in the visible responses. I’m not sure how to feel about this. I do like the quality of the responses though.
Anyone else with similar experience? What contents triggers it for you?