Is Claude Making You Dumber?

If you feel smarter every time you use Claude, you might be getting dumber.

I was.

You're working on something and you just want a quick answer. "Claude will just figure it out."

You're treating Claude like a vending machine. That reflex, offloading your mental effort, is the trap. The path of least resistance becomes the only path you know.

That's what I'm fighting against.

I realized I was blindly trusting Claude's responses when I started reading the sources it was citing. I was researching an MIT study using zurf. When I told Claude to research a question I had, it usually used the ask command, which internally uses Perplexity. It turns out these responses were usually hallucinations from Perplexity, even when it "grounds" its responses from the internet. Claude was blindly trusting Perplexity, and I was blindly trusting Claude.

So I told Claude to fetch from multiple sources directly and compare all of them to give me a final response. And I found a different issue. The sources Claude was using were blog posts, third-party websites, middlemen, not the real source, the research. When I read one of the articles and compared it with the content from the real source, I found that the interpretation of the results in the blog post was different from my interpretation.

The brain optimizes for efficiency. Remove friction, and it stops thinking. An MIT study, "Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task", found that LLMs reduce friction in answering questions. However, this convenience comes with a cognitive cost. We tend to accept whatever LLMs tell us without critically evaluating it. The path of least resistance with AI leads to cognitive disengagement.

But, there is a better way.

28 questions. 460-line decisions file. 10 research files. Because I asked Claude to "grill" me about a feature I thought I'd already figured out.

I have a "grill-me" skill that relentlessly asks you questions until you and Claude reach a shared understanding of the idea, plan, or design. Claude automatically creates a "decisions" file where we put context, each question with exploration and final decision, pending researches, and finally a short paragraph explaining the why.

I wanted to build the subscriptions feature for Resumeskit. I had an idea in my head about what I wanted, and I thought it was enough. But I decided to use the skill in case I missed something.

The feature mutated into the most complex one in the project. Now I better understand it, and this document helps me and Claude to build a better version of it. It's not perfect, but it's way better than a 10-line prompt explaining the feature.

This grilling process requires a lot of mental engagement. It's an exhausting process, and I believe it should be. That's where I spend more time now. On thinking, and using Claude to help me think.

The AI that saves you mental effort makes you dumber. The AI that creates mental effort for you makes you sharper.

Go to the source. When researching something, go deep. Use Claude to give you diversity, but go to the source. Read it. Learn from it. This takes time. But this beats just accepting whatever Claude says.

Going deep is where I find great books/material to read and learn from, like "Thinking in Systems" by Donella Meadows, which I decided to read while I was researching for Stop and Think: Solve It Once, Use It Forever. Now I see everything as a system.

Make Claude grill you. Use Claude to interview you about any idea, plan or design you have. Use it to find the gaps, the edge cases, to define pros vs cons. It will force you to think clearer. You'll be surprised to know the things you haven't questioned.

Follow your curiosity thread. Be curious, ask questions. That's how you learn. It's so easy to learn new concepts with Claude, but you have to do the mental work. Don't leave everything to Claude. What started as "Why is JavaScript so popular for building software products when there are frameworks like Laravel on PHP that are way better?" ended with me exploring Python and FastAPI, because most AI-first businesses run on it.

Are you going to leave Claude to do the thinking for you? Or are you going to use it to make you sharper?