I ran it through ChatGPT

“I just ran it through ChatGPT,” my teammate said, “and this came out.” An online presentation followed. There were no questions.

In the Andrej Karpathy video I shared earlier this quote came up: “You can outsource your thinking, but not your understanding.” I think that’s too optimistic. Human nature is just to optimize for laziness.

AI makes mistakes all the time, or even lies to you, or makes up facts, or is based on outdated information. It’s constantly in the news, and everyone knows it. AI cites an article about a completely fabricated disease. Lawyers showing up with nonsense sources. And yet everyone keeps asking it everything, uncritically, and believing the answer. Because it sounds like a human.

The great cop out of 2026

“I ran it through ChatGPT” is offloading responsibility. That’s what happened with that presentation. “I just ran it through ChatGPT, and this is what it said.” Then nobody asks any questions. Because what is there to ask about? Everything happened in the black box. The team member probably didn’t even know what assumptions went in, so you can’t quiz them either. This is what is now being called ‘cognitive surrender’. Little thought was given.

“I ran it through ChatGPT” is therefore more than a conversation-killer. Normal conversation-killers (“it is what it is”, “this is just how we do it”) are recognizably empty: you know that someone doesn’t want to talk about it because they don’t really know. Which you can then address: Is that really so? Why do we always do it that way? But the outcome of “I ran it through ChatGPT” is worse. It looks like a worked-out plan. Tables, charts, conclusions. Looks like proper human work. But in the conversation it’s an endpoint. Because there’s nothing to ask. Ok. Thanks.

This is not a plea against AI by the way, I use it every day myself. But it is a plea for understanding.

There are two problems here. One is that AI gets too much room to make mistakes. The other sits in how the outcomes get discussed.

How to get AI to do better work

• Ask for sources. And then actually check them. This cuts straight through 90% of cases. I had a question about housing benefits where AI confidently gave the wrong answer, and when I asked for the legal articles it pointed me to ones that were about something completely different. After 5 rounds back and forth, ChatGPT finally admitted: “You’re right. My earlier claim (‘family is not allowed’) was incorrect.”
• Spot the vague phrasing. “This is common” is one LLMs love to use. “Do you have a source for that?” is the right reply.
• Build your solution step by step, and make sure to check and understand each step. This is often easier than to have it built the final product, which is much more difficult to verify.
• Ask for a different route, or start a new chat where you approach the question completely differently.
• Understand it, and find a way to verify the outcome. First principles. Does this even make sense? What if I put that to 0; what outcome do I then expect, and do I get that?
• Test, debug, test, debug. AI is only interested in the happy path.

How to discuss the outcome better

• Show the assumptions in the input. If you don’t know them, back to square one, until you do.
• Show what you didn’t check. “This is an assumption, but I couldn’t verify it.”
• Describe your reasoning before you share. If you can’t explain it, you don’t understand it well enough yet.
• Take responsibility: This is my story. AI was a tool, but I’m the owner and I’m responsible for the outcomes. Ask me anything.
• And as listener/reader: Ask them anything. There’s social discomfort in asking hard questions (“I don’t want to embarrass them”) which makes some people avoid it. Do it anyway. Maybe with a “let me think along with you for a second” or “have you compared this outcome with xyz?”

A good test: would I tell my best friends or dearest family member this conclusion with 100% confidence, or let them use this app? My home battery calculation is a good example. Would I let my dad spend hundreds, even 1000+ euros, based on a wrong conclusion? Every time I noticed something I couldn’t quite understand I asked Claude about it, and Claude would find that there was a small mistake somewhere, or something hadn’t been included, or something had been misinterpreted. And the numbers would shift again. Only when I could see and understand exactly how the charging strategy went day to day did I have confidence, and only then did it feel like “I made this, with AI as a tool.”

Outsource the work, but not the understanding. Tempting as it is.