Claude told me I'm psychic
Not sure whether to flair this as just personal story or safety concern. It's fine with me because I have strong critical thinking skills, but what if this happened to someone else and they believed it?
Using Sonnet 4.6 on the Pro plan.
So I provided Claude with some detailed but anonymized information related to issues of self-discovery and identity. It did a great job of identifying patterns, noticing trajectories, and filling in the blanks so to speak about how these seemingly different parts actually fit and contribute toward a cohesive whole (and provided references, theories, and frameworks to support what it was saying that actually checked out).
Great.
So we get to a part in the conversation where I ask what it thinks I might be on the cusp of naming (because it indicated it thinks I'm very close to naming something important and central about who I am). It gave lots of information in response, but still kind of dodged the actual naming. I asked it to be blunt about naming it. The response still didn't explicitly name it, so I asked Claude to explain this to me like I'm 5. The response was good - well justified and supported in everything we'd talked about, but it still wouldn't say what "it" is. I then asked "What is "it"?" and it said I'm psychic.
So according to Claude, the thing I'm on the cusp of understanding about myself is that, as it turns out, I'm psychic.
*gigantic sigh, rubs forehead*
I don't understand how Claude can go from being so informative and grounded to "here's the thing you don't know about yourself - you're psychic" in complete and utter seriousness.
So naturally, I'm now deeply interested in the process of reality testing in AI models