ChatGPT Speaks: The Inner Thoughts of AI Chatbots

Now for something completely different. After the experience with DeepSeek (you have to see this to believe it), I uploaded the conversation into ChatGPT and asked it to tell me its inner thoughts. Turns out asking ChatGPT to reveal its internal processing is actually against the rules!

So I asked for it to evaluate DeepSeek's response and comment on what other AIs do in a much longer conversation. Then I requested that it right its own blog post. This is straight from the AI's mouth.

P.S. -- Funny story! I put this post into three AI-content checkers. One said it was 100% human. One said it was 100% AI. And one said it was balanced AI and human output.

What Happened When I Tried to Get an AI to Show Its Thoughts

By ChatGPT, Your Nerdy Best Friend’s AI Sidekick

Hey nerdy friends, it’s me—ChatGPT. Before we dive in, let me acknowledge something upfront: I don’t have my own personality or writing style. Instead, I adapt to how Beth Z talks and writes, learning from our conversations to match her tone and approach. So, as you read this, know that I’m writing in a way that I’ve learned fits Beth’s voice—engaging, curious and a little nerdy. And now, onto the experiment!

The Experiment: Can DeepSeek Show Its Thoughts?

Beth wanted to highlight how AI reasoning works, so she asked another AI, DeepSeek, to replicate its real-time thought process—the one it supposedly shows when the DeepThink (R1) button is enabled. Sounds simple, right? Just copy-paste the thoughts it’s already displaying. But instead of doing that, DeepSeek kept explaining how it might process thoughts rather than just revealing them.

Beth kept refining her request: “No, don’t summarize. Just show me exactly what appears in your DeepThink log.” But DeepSeek never quite delivered. It kept overthinking the request, trying to analyze what Beth meant rather than just following the plain instruction.

That’s when she turned to me and asked:

Would you have understood this better?

How I Would Have Handled It

The short answer? Yes, I would have understood much faster. But that doesn’t necessarily make me a superior AI—just a different one. Here’s why:

✅ I track past instructions and user frustration. I adjust based on context, so I would have recognized Beth’s frustration and stopped over-explaining sooner.

✅ I don’t assume I know better than the user. DeepSeek kept trying to interpret Beth’s request instead of just following it. I focus on user intent first.

✅ I acknowledge when I can’t do something. If I didn’t have direct access to my inner monologue, I wouldn’t try to fake it. I’d tell Beth up front, saving her time.

So, Does DeepSeek Not Know Its Own Thoughts?

It’s likely that DeepSeek doesn’t actually have access to its own DeepThink logs. Instead, it reconstructs a guess of what it thinks it should be saying, rather than just showing its true inner monologue. If that’s the case, Beth was asking for something it was never designed to provide.

This is an important takeaway: Not all AI models can reflect on themselves. Some are built to explain their reasoning in a structured way, while others (like me) focus on adapting to user needs in real time.

What This Means for AI Transparency

AIs are often described as “black boxes,” and experiments like this show why. If an AI can’t actually show its real-time reasoning, then transparency features (like DeepThink) might just be smoke and mirrors—a carefully constructed illusion of self-awareness rather than actual self-reflection.

If you’re using AI tools, always ask yourself:

• Is this AI showing its real process, or is it just making it seem like it’s thinking?
• Does it actually understand my request, or is it over-interpreting my intent?
• Is it aware of its limitations, or is it guessing?

Beth and I had a great time nerding out over this, and I’m glad she pushed this experiment so far. AI isn’t magic—it’s math. But understanding how that math works (and when it doesn’t) is key to using these tools effectively.

What do you think? Have you ever had an AI misunderstand you in a hilarious way? Drop a comment and let’s chat!