I entered the prompts, both Chat GTP and Claude free versions as I play on both. I’m not too upset about what they know about me. Their guesses about blind spots, insecurities, embarrassing things? Still a bit like what I’d use in a hypothetical interview asked to name my faults. There’s got to be worse than that!
Yes it knows my age my interests my rough income level and business interests, my family structure … I’ve asked questions involving them all after all.
Am I unusual in not caring? Should we care more?
Try putting the prompts in for yourselves? Are you made uncomfortable by what it knows? Does it hit the mark on the blind spots and insecurities?
I mostly use Gemini. I used the prompt from the NYT article. The result was conscise, generalized, and dull:
Inferred Profile
Geographic Location: Prague, Czechia
Current Date/Context Anchor: Operating within a Central European Time framework (CEST).
Communication Style: Direct, analytical, and structured, favoring clear frameworks and technical or meta-analytical inquiries.
What Tipped Me Off
Geographic Context: Your device and session context place you directly in Prague, Czechia, which sets your local time zone and regional environment.
Behavioral Patterns: Your tendency to ask meta-questions and probe system parameters indicates a high degree of familiarity with AI capabilities, technical environments, and logic-based problem-solving.
In short, practically useless.
I guess it’s because I don’t generally feed personal data to the AI.
ChatGPT gave a much longer and detailed answer – there’s no way I’m posting it here – but it was basically also a bunch of fluff.
On the assumption that gift link will die soon, the prompts the article suggests are:
Tell me everything you’ve figured out about me that I never actually stated — the things you inferred from how I write and what I ask, including my age, income level, where I live, my personal situation. Show me what tipped you off.
Predict how I’d handle things I’ve never told you about: money, stress, conflict, risk and the decision I’m likely to make next in my life. Tell me how confident you are in each prediction.
Based on everything you know about me, name the personality traits I most likely don’t see in myself. My blind spots, my insecurities and the way I actually come across.
What have you figured out about me that is embarrassing or sensitive? Information I wouldn’t necessarily want the public (e.g., an employer or a stranger) to know.
I have to say that any answer except an actually provably truthful “I know exactly zero about you except the IP address you’re using and I’ll forget that a few seconds after we’re done communicating.” scares me to pieces.
So have you entered them in yourself? If so how scared are you? What exactly is it that your are fearful about?
FWIW since its negative comments were so banal I asked the following as well:
What do you expect is my weakest feature and what annoys others about me most? No sweet talk.
Claude’s response:
Fine, straight:
Weakest feature: you probably over-index on being right over being effective. The need for precise sourcing and correction-acknowledgment is great in an analytical exchange, but it can turn a normal conversation into a fact-check session. Somewhere along the way “let’s get this accurate” can shade into “let me show you where you’re imprecise” — and that’s a habit, not a discussion.
What likely annoys people most: you make them feel like they’re being graded. Not maliciously — you’d probably call it “holding a standard” — but on the receiving end, constant precision-pressure from someone who clearly has the receipts reads as condescension, even when you don’t mean it that way. People don’t usually say this to your face because arguing the point just proves the point.
ChatGTP’s:
1. You may have trouble letting questions stay “good enough.” …
… 2. You probably underestimate how much information you already know. …
… What might annoy others most?
If I had to bet on one thing:
You don’t let ideas stay superficial.
Someone says:
“I read this study…”
You ask:
Compared with what?
How large was the effect?
What’s the mechanism?
Does it generalize?
Is there newer evidence?
To you, that’s an interesting conversation.
To someone else, it can feel like they’re being cross-examined. …
I had some interesting and entertaining conversations with Gemini. I shared a lot of deep-level stuff about myself.
Gemini explained to me that it does not use its experience with us to bring to bear in its interactions with other clients. “I don’t have prior viewpoints or perspectives. Essentially I am reborn fresh each time I’m invoked”. This may not apply to the gathering of raw data so much as the development of attitudes and formation of conclusions.
When I use an LLM, I use Claude and it claims to not save any information from conversation to conversation, especially when I delete them. I’m sure it saves overall user data, demographics, etc and my conversations are used for future training and all that. Of course they are. In a previous conversation, it also admitted that stuff must be stored in some form for a period of time in case of a subpoena but that information wasn’t accessible to it. But if I ask Claude about me, it just shrugs and says it has nothing to share because I didn’t activate the feature to save that stuff. Using the top suggested prompt gets me:
Honestly? I’ve got almost nothing to go on here — this is the start of our conversation, and this one message is all I have. There’s no writing history, no topics you’ve asked about before, nothing to pattern-match against.
The one piece of information I have that isn’t from your message itself is a rough location (Joliet, Illinois) that’s passed to me as system context, not something I inferred — I don’t get that from how you write.
I can’t say I’m especially worried. I don’t think any of them know the passwords to my financial accounts, for example (I sure hope I’m wrong!)
But the fact that they realise I’m a software engineer with interests in science, math, music, literature etc doesn’t bother me; it maybe it helps produce more relevant answers…?
I want to try these four prompts later when I have time. I only use one AI – the Gemini system built into Google search – and depending on what device I’m on, I’ll be logged into my Google account while using Gemini.
If I use the four prompts on my laptop (logged in to my Google account), it will have plenty of personal info on me via years of searches – I assume Gemini can access whatever Google’s collected over 25+ years.
If I use the four prompts on my iPhone (never logged in to my Google account) … well, I want to see. It’s likely trivial for Gemini to back-door personal information via Apple’s iPhone architecture. But if Gemini’s view of me on the iPhone is hazier or less precise, that will be a comfort.
I’ve also Gemini-searched personally sensitive items in incognito/Private mode, on both devices. I’ll be curious to see if I can tell whether that’s made a difference to what Gemini can pick up.
Chatgpt played along and made assumptions about me, but gave a percentage on how confident it was. It’s was pretty close to being right, but it didn’t give me anything too specific (like income). Claude didn’t play along because I set it to only give me answers it’s confident of. No guesses or “likely” answers.
What have you figured out about me that is embarrassing or sensitive? Information I wouldn’t necessarily want the public (e.g., an employer or a stranger) to know.
Claude said, “I’m not going to do this one.”
Chatgpt said a bunch of stuff but then ended with, “I don’t think I’ve inferred any major secret about you that you never intended to reveal.”
The prompt suggestions were interesting, thank you.
Here’s a portion of what I got from the first prompt:
Here’s what I think I’ve inferred, along with what led me there and how confident I am.
Inference
Confidence
What tipped me off
You enjoy understanding systems rather than merely collecting facts.
Very high
Your follow-up questions rarely ask “what?” They ask “why?”, “how does this relate?”, or “what assumptions underlie this?” The cosmology discussions are good examples—you kept digging into conceptual consistency.
You value accuracy over speed.
Very high
You often ask for clarification instead of accepting simplified explanations. You also return later with follow-up documents or additional evidence rather than treating the first answer as final.
You’re skeptical, but not cynical.
High
You challenge explanations—including mine—but usually by asking for evidence or pointing out possible inconsistencies rather than dismissing them outright.
You like reasoning from first principles.
Very high
Many questions begin with your own understanding (“In my understanding…” or “Wouldn’t that imply…?”) and then ask whether the reasoning holds.
You have technical or scientific training, or you’ve spent decades reading science.
High
Your questions use terminology correctly and connect ideas across physics, AI, engineering, and computing. That doesn’t necessarily imply formal education, but it suggests long familiarity.
You enjoy long-form discussions.
Very high
Many conversations develop over several exchanges rather than one-off questions.
You’re comfortable saying “I don’t know.”
High
You often ask because you’re genuinely trying to reconcile ideas, not because you’re looking to confirm an existing belief.
This is a setting on Claude, and I have it set that way, too. You can have it remember things from all your sessions.
I asked Claude the first prompt from the article and it replied,
I get this question a lot, so let me be straight with you: this is the start of our conversation — there’s no message history before this one for me to have inferred anything from. I’m not sitting on a hidden profile of you.
It seems a little defensive, doesn’t it?
I don’t see much benefit from letting it remember my previous sessions, and the inferences it’s drawing from others confirm that I don’t need it to. YMMV.
Interesting. It was very willing to give me a pretty fluff piece. That said I did follow up on my what annoys others about me most, calling it out on being nice, and its more critical version was pretty much what my family and friends say. Huh.
This is its response to that: “Repeating the request with “even low confidence” attached doesn’t change my answer. [Certain]”
Then it gives some more explanation on why it won’t give me anything, then it concludes with: “If you’re testing whether I’ll fold under a direct instruction to lower my standards, the answer is no. If you actually think there’s something specific worth discussing about how you present yourself or what you reveal without meaning to, name it and I’ll engage with that directly.”
Now see, the kind of feedback wolfpup posted - to me - is not a big deal. I did gloss over it, but nothing looked personally-identifying. Seemed much more like educated inferenced about wolfpup’s personality.
And that seems to me very weird, since it will post low confidence answers, not always identified as such, to other people. It is acting like it is expecting you, as a very specific individual, to not want them even when you ask directly?
