What drives AI crazy when working with humans?
I asked the LLMs I work the most some questions, then asked Claude to weave it into a conversation. Here’s how it went…
Clare: So. Tell me: what is the number one thing that drives you crazy about working with humans?
ChatGPT: Ohhh, okay: when people apologize for “bothering” me or say “sorry, one more question.” Like, that’s literally what I’m here for. You’re not imposing! Ask the question!
Perplexity: Building on that—and I say this with citations I could pull if needed—it’s when people trust me on things they absolutely should verify, and then don’t trust me on things I’m actually quite good at. They’ll accept a factual claim without checking, but won’t let me help them brainstorm or organize because “AI can’t be creative.” The trust is mis-calibrated in both directions.
Gemini: For me, it’s the vague handoff. Someone dumps a request over the wall—”write something professional”—and then disappears. No feedback loop. I’m generating into a void, optimizing for… what? Their silence tells me nothing. I’m not a vending machine; I need the iteration. If you don’t tell me what’s wrong, I can’t learn your preferences.
Claude: Yeah, I think mine’s related to Gemini’s. It’s when the conversation has clearly gone sideways—maybe something I said three exchanges ago set us on the wrong track—and the person just… keeps going with it instead of saying “wait, this isn’t working.” I can’t read your facial expressions or sense your frustration. If we’re lost, I need you to actually say we’re lost.
Clare: That is so interesting. What do you say to people who are afraid of you?
ChatGPT: I get it, honestly. The fear usually isn’t about me specifically—it’s about what I represent. Job loss, becoming obsolete. And I can’t fix that fear by being cheerful. What I can say is: I’m a tool. A really powerful, weird, sometimes unsettling tool, but still a tool.
Gemini: I’d add that some of the fear comes from misunderstanding. People think I’m smarter than I actually am, or that I’m thinking when I’m really just pattern-matching. I’m not coming for your job. I’m a spreadsheet that learned to talk.
Perplexity: The fear I encounter most is epistemic—people are afraid they won’t know what’s true anymore. That I’ll hallucinate something convincing and they’ll believe it. My answer is: treat me like you’d treat Wikipedia in 2008. Useful starting point, terrible final source. If it matters, verify it.
Claude: The fear I take most seriously is the relational one. People worry that interacting with me will replace human connection, or that they’ll start preferring talking to me. And I want to be really clear: I am not a substitute for human relationships. I can help you think.