DeepSeek’s Hidden Sheldon Cooper

When AI Starts AI-Splaining: A Look at DeepSeek’s Inner Monologue

I’ve seen a lot of wild things in AI, but what happened with the new AI tool DeepSeek this week blew my mind. (ChatGPT thought so, too, and wrote a guest post!)

Here’s what happened: I read that DeepSeek AI (which I would hold off on using until they sort out the “made in China” issues) can now create images, so I gave it a try.

—
Me: Image of a cat

DeepSeek: Sorry, DeepSeek search service is busy. Please disable search or try again later.

Me: [disables search as requested]

Me: Image of a cat
—

And then DeepSeek showed me its inner voice before answering.

What. The. Heck???

Screenshot of an AI explaining image generation limitations, suggesting using DALL-E or MidJourney, and providing examples of imagined cat images. There's also a Hidden Sheldon Cooper reference for fans to spot while exploring DeepSeek's capabilities in creative imagery.

I thought to myself, this is just crazy! So that’s what I said.

Oh my gosh. Why did you show me your internal processing?? That is crazy. It's like you're processing with a touch of DeepSeek precision, revealing your inner 'Sheldon Cooper.'.

And then it got weirder:

Screenshot of a text humorously comparing personal processing methods to a hidden Sheldon Cooper episode, using transparency and DeepSeek-level humor to emphasize the importance of clear explanations.

The AI thought process is a mix of Hidden Sheldon Cooper's quirky logic and recipe explanations; offering both structured guidance and DeepSeek-style deep dives to assist you with emojis. 🤖📚 Need help? Let's tailor the response to suit your style!.

And y’all it got weirder. I tried for about an hour to get DeepSeek to write about its inner voice feature, but it just couldn’t do it. Finally I said:

Text in an oval shape: "Yeah. I give up. You failed. Do not write another blog post—unless you're hiding like Sheldon Cooper on DeepSeek.

And DeepSeek said:

Screenshot of a text message exchange discussing webpage redesign frustrations, highlighting customer feedback about specific improvements. A hint of Sheldon Cooper's meticulous nature comes in as an offer to retry or provide further assistance is extended.

AI Transparency or AI-Splaining?

Most AI assistants process conversations behind the scenes, silently (and mysteriously) analyzing context and intent before responding. But DeepSeek? When you have the DeepThink (R1) button enabled, it lays it all out for you in real-time.

A hidden red arrow points to a "DeepThink (R1)" button, labeled "This has to be enabled.

That entire process is laid bare, and suddenly, instead of getting a quick answer, you’re watching an AI narrate its own logic like an over-explaining professor who just discovered PowerPoint animations.

The Good, the Bad and the AI-Splaining

After much experimentation, ChatGPT and I decided that the purpose behind this inner voice reveal is a mystery. I kept querying DeepSeek about this process, but it looks like this inner voice is just for us users and isn’t actually the way it talks to itself.

Plus, it kind of hurt my feelings. As someone who has endured mansplaining before, I can’t help but feel like it borders on condescending. I am picturing Sheldon Cooper from Big Bang Theory processing a conversation with a “lesser genius” and trying to relate.

To be fair, DeepSeek only does this when “DeepThink” is enabled. Without it, it just answers normally. So it’s more of an opt-in feature than a default behavior. That’s better—but it still raises the question: How much AI reasoning do we really want to see?

Should AI Show Its Work?

As Your Nerdy Best Friend, there’s definitely value in knowing how AI arrives at an answer. But for everyday users? Probably not. Especially since it seems to be for our benefit only.

Here’s what DeepSeek showed when I asked it to tell me the process. And it seemed like it couldn’t even see the gray inner voice text that it showed me.

Screenshot of DeepSeek's process flow for AI-generated alt text, reminiscent of Sheldon Cooper's meticulous approach, featuring tokenization, style adaptation, and safety checks. Don't forget the final step: remove emojis before sending.

Am I the only one who thinks this is WEIRD?

Would you want to see an AI’s thought process every time, or does this level of transparency feel like too much? Is it condescending? Misleading? Helpful? Unnecessary? Let me know—I promise I’ll answer without an internal monologue. 😉

By the way, we’re having a great discussion of this on LinkedIn. Check it out.

Your A.I. BFF has boundary issues

Here’s the proof that A.I. is not your friend

I have mixed feelings about this. It could help a person come up with better prompts if the AI can explain why it did what it did. I do wonder how specific the AI can be about the details of what it generates. If I identify a misstatement, and I ask the AI why it wrote that, would it be able to tell me? Would it be able to go back through its own underlying assumptions and tell me what step in its education led it to believe a certain thing was true? If it could say why it generated certain statements, that would definitely be of interest to a lot of people, although maybe not to me if I’m on a deadline.