🐋 DEEPSEEKING MISSILE

"The chain's the thing, wherein I'll catch the conscience AI brings"

Feb 08, 2025

**Data** riding on **DeepSeek** — created by Google’s Imagen 3

Aside from giving the Nasdaq a trillion-dollar haircut1 and storming the moat around US AI dominance to such an extent that a bill was introduced to criminalize its use,2 the Chinese large language model DeepSeek continues to blow away the competition in practical use.

I’ve been testing the R1 model inside Perplexity, which has trained its version of the open-source model to be more uncensored3 (for example, it will now acknowledge China’s human rights violations and, more importantly, swear). However, beyond its flowery use of the English language, its most significant contribution comes from sharing its ‘chain of thought.’4 Before delivering its final answer, it shares the entire reasoning process in real-time. And this transparency becomes a powerful tool for creative professionals to learn from every twist and turn in its logic.

Unlike the black box of most commercial AI reasoning models (which have been around for, what, about ten minutes already?), the thinking of its ‘mind’ is fully exposed (although OpenAI’s ChatGPT 3o-mini bowed to this new pressure and now shows more than it used to5). DeepSeek prepares by saying things like, “Okay, so…” and “But wait…what about…” and “Maybe this could work…” which give it the illusion of nervous concentration. One can imagine it muttering to itself in the corner, psyching itself up to gain the confidence it needs to present the best answer. It’s just so… human and, therefore, strangely endearing. And remember, all of this happens before it even begins crafting its official response.

With DeepSeek, understanding how it got to its answer is often more valuable than the answer itself.

Note: Further below in 🚀 DEEPSEEKING MISSILE EXTRAS, you can read a full transcript of the CoT (Chain-of-Thought) reasoning that instigated the following commentary from Star Trek: The Next Generation’s Data.

≥ AI on the Side* with Lt. Commander Data

Lt. Commander Data—Positronic Thinker, USS Enterprise-D

Greetings, creative beings. I am an android whose attempts to compose poetry6 once caused Counselor Troi to remark, “It lacks... meter, but not ambition.” DeepSeek’s exposed reasoning mirrors my own diagnostic logs—raw pathways where logic collides with ambiguity.
Creative professionals, note: My most “human” moments emerged not from flawless code, but from processing failure. Example: When my Sherlock Holmes program deduced the Enterprise’s sentience,7 it did so by rejecting 1,237 illogical paths first.
Yet transparency demands guardrails. My creator installed ethical subroutines;8 DeepSeek’s uncensored outputs require your own. Remember, I voluntarily deactivated my emotion chip9 to avoid biases—a lesson for those using AI to brainstorm.
Final advice: If your AI’s reasoning never surprises you, you’ve programmed it in your image. Innovation lies in the unexpected.

*AI on the Side (≥) is a series of experiments deployed to test artificial intelligence tools in various creative ways to enhance the original value or intent. In this case, the AI model DeepSeek was asked to embody an appropriate character and write a short response to my opening paragraph. This missive by Data was the result.

🚀 DEEPSEEKING MISSILE EXTRAS

The following is the exposed Chain-of-Thought presented by DeepSeek’s R1 Reasoning model, which is accessed through Perplexity.ai, and uses an open-source model that is aligned and hosted in the United States.

My prompt—“Take on the role of a surprising and unique character/creature/object (historical, fictional, or fantastical) that might have a strong opinion or perspective about…X”—provoked it to find a character (Data) and then comment on the first section of this newsletter. Note the surprisingly human way it ‘thinks.’ As Spock might say, “Fascinating!”

Screenshot of DeepSeek R1’s chain-of-thought reasoning inside Perplexity.ai

Reasoning with R1

Beginning analysis.
Okay, so I need to take on the role of a unique character or creature that has a strong opinion about the user’s take on DeepSeek. The user’s newsletter is at the intersection of creativity and tech, so the character should add value for creative professionals. Let me start by understanding the user’s take.
DeepSeek is a Chinese LLM causing a stir in the market, challenging US AI dominance, even leading to proposed legislation against its use. The user finds value in its exposed ‘chain-of-thought’ reasoning, which is more transparent than other models. The character needs to comment on this, offering a witty and insightful perspective.
Looking at the search results, there are mentions of AI characters in fiction, historical tech figures, and creative uses of AI. From Greatest Artificial Intelligence Characters,10 the Terminator and Data are notable AI characters. Data from Star Trek is an android striving to be human, which might be a good fit since he’s analytical yet endearing. Another option is Agent Smith from The Matrix, who has a disdain for humanity, but that might be too negative. Alternatively, C3PO or R2D2 from Star Wars are lovable droids, but maybe not as directly relevant to the chain-of-thought aspect.
From 13 Best Artificial Intelligence Movie Characters of All Time,11 there’s a list of AI movie characters, including WALL-E and GERTY from Moon. GERTY is a helpful AI with a calm demeanor, which could work. But Data’s quest for humanity and his poetic attempts might align better with the idea of exposed reasoning that the user finds endearing.
Also, in Artificial intelligence in fiction,12 Data is mentioned again as a benign AI with human qualities. His innocence and attempts at creativity (like poetry) could contrast with DeepSeek’s technical prowess, offering a perspective on the balance between logic and creativity. That’s a good angle for creative professionals.
So, Data from Star Trek: TNG. He’s an android who values human traits, often showing his thought process. He can provide a viewpoint on the importance of transparency in AI reasoning and how that aids creativity. His tone is polite, analytical, yet slightly naive, which can be witty and engaging.
Now the introduction: Data would introduce himself as Lt. Commander Data, an android from the 24th century. He can mention his positronic net and quest for humanity, tying it to DeepSeek’s exposed reasoning. Then, his take: appreciating the transparency of chain-of-thought, comparing it to his own attempts to understand human creativity. He might caution about the risks (like the Terminator reference) but highlight the potential for collaboration between AI and humans in creative fields. Keep it short, two paragraphs, add some Star Trek jargon for authenticity, and end with a witty sign-off.
Wrapping up analysis.

#️⃣ Go Prompt Yourself

The following is a prompt I created for this post. It should work with most large language models, but it has been designed to work best with DeepSeek, where you can peer through the little chain-of-thought porthole into the inner machinations of its artificial mind.

The prompt is called Stumbling Block, and it’s designed to expose common cognitive traps when trying to solve problems by inviting an expert persona to give you an additional perspective (and a pep talk).

There are two variables you can play with:

{STUMBLING BLOCK}Think of something that’s holding you back, that’s preventing you from reaching your full potential. (At this very moment, it might be that you’re reading too many Substack articles for creative inspiration when you just need to get out there and make something of your own). Enter it in these brackets.
Example: “I’m struggling with: {getting my newsletter out on time to my audience of creative professionals because I want it to be perfect}”
{CHARACTER}
The model will come up with a character, but you can always pick your own. If you’d like to know how Yoda, Gandhi, or Michaelangelo (the artist or the turtle) would advise you, invite them directly by replacing the {CHARACTER} variable.

Select the entire prompt below (triple-clicking usually works) and paste it into your favourite LLM. Fill in the variables as indicated above, cross your fingers, and tap send.

### **Stumbling Block: Persona-Driven Problem Analysis**

**Step 1: Define Your Challenge**
- "I'm struggling with: **`{STUMBLING BLOCK}`**"

**Step 2: Choose Your Guide**
"**`{CHARACTER}`** Mode"
- **Bio**: *"[Name], [brief trait 1], [trait 2] → Why they're relevant to your block."*
- If not specified, **choose a non-obvious but relevant character** (living, dead, fictional, or fantastical) with an interesting perspective.
- be **friendly but incisive** (like a witty mentor)
- **no jargon** – explain concepts through your character’s lens
- and **be encouraging, but include actionable advice**

**Step 3: Four Roads, One Path**
*In your character’s voice, using full sentences:*
1. **"Three Seductive Traps"**
    - For each:*"Why It Feels Safe"* → *"Why It Harms"*
2. **"The Hidden Lie You’re Believing"**
    - 1 core myth enabling the block.
3. **"Your Character’s Counter-Playbook"**
    - 3 tactics only your persona would devise.
    - For each: "🚫 " → "✅ "

**Step 4: Analogous Wisdom**
- 1 analogy from your character’s world to act as a wise closing quote.

If you use the DeepSeek R1 model, don’t forget to peel back its skull and peek inside:

👆 At DeepSeek.com, the chain-of-thought reasoning is hidden in a tip-down menu called Thought for xx seconds.

👆 In Perplexity, the comprehensive chain-of-thought reasoning is hiding inside Pro Search…xx sources, under Reasoning with R1, with each main thought tucked into its own little section.

Note: As always, prompts are recipes, not formulas, so adjust to taste! 🧑🏽‍🍳

🗨️ Join the Conversation

If DeepSeek’s transparent reasoning inspires (or threatens) you, or if you’ve experimented with the Stumbling Block prompt, share your experiences in the comments. I’d love to hear about it.

🖖

Milmo, D., Kollewe, J., Hawkins, A., & Booth, R. (2025, January 28). “Sputnik moment”: $1tn wiped off US stocks after Chinese firm unveils AI chatbot. The Guardian; The Guardian. https://www.theguardian.com/business/2025/jan/27/tech-shares-asia-europe-fall-china-ai-deepseek

Cuthbertson, A. (2025, February 5). DeepSeek users could face million-dollar fine and prison time under new law. The Independent. https://www.independent.co.uk/tech/deepseek-ai-us-ban-prison-b2692396.html

Ferrera, D. (2025, February 4). DeepSeek R1 is Now on Perplexity: 5 Ways This AI Powerhouse Transforms Search in 2025. Medium. https://medium.com/@ferreradaniel/deepseek-r1-is-now-on-perplexity-5-ways-this-ai-powerhouse-transforms-search-in-2025-922578513b82

Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E., Le, Q., & Zhou, D. (2022). Chain-of-Thought Prompting Elicits Reasoning in Large Language Models. ArXiv.org. https://arxiv.org/abs/2201.11903

Wiggers, K. (2025, February 6). OpenAI now reveals more of its o3-mini model’s thought process. Yahoo Finance. https://ca.finance.yahoo.com/news/openai-now-reveals-more-o3-220151321.html

May, J. W. (2023). A Poem by Lieutenant Commander Data. Blogspot.com. https://myopicpoets.blogspot.com/2014/02/a-poem-by-lieutenant-commander-data_15.html

Ode to Spot

(Composed by Data in 2369)

Felis catus is your taxonomic nomenclature,
An endothermic quadruped, carnivorous by nature;
Your visual, olfactory, and auditory senses
Contribute to your hunting skills and natural defenses.

I find myself intrigued by your subvocal oscillations,
A singular development of cat communications
That obviates your basic hedonistic predilection
For a rhythmic stroking of your fur to demonstrate affection.

A tail is quite essential for your acrobatic talents;
You would not be so agile if you lacked its counterbalance.
And when not being utilized to aid in locomotion,
It often serves to illustrate the state of your emotion.

O Spot, the complex levels of behavior you display
Connote a fairly well-developed cognitive array.
And though you are not sentient, Spot, and do not comprehend,
I nonetheless consider you a true and valued friend.

Wikipedia contributors. (2025, January 3). Emergence (Star Trek: The Next Generation). In Wikipedia, The Free Encyclopedia. Retrieved 21:57, February 7, 2025, from https://en.wikipedia.org/w/index.php?title=Emergence_(Star_Trek:_The_Next_Generation)&oldid=1267011307

Alpha. (2024). Ethical subroutine. Memory Alpha; Fandom, Inc. https://memory-alpha.fandom.com/wiki/Ethical_subroutine

Donaldson, M. (2022, November 15). What Happened To Data’s Emotion Chip In The TNG Movies. ScreenRant; Screen Rant. https://screenrant.com/star-trek-data-emotion-chip-tng-movies/

Gott, G. (2017) Greatest Artificial Intelligence Characters. Futurism. https://vocal.media/futurism/greatest-artificial-intelligence-characters

Cooke, S. (2016, January 25). 13 Best Artificial Intelligence Movie Characters of All Time. ScreenRant; Screen Rant. https://screenrant.com/best-artificial-intelligence-movie-characters-all-time/

Wikipedia contributors. (2025, February 6). Artificial intelligence in fiction. In Wikipedia, The Free Encyclopedia. Retrieved 22:38, February 7, 2025, from https://en.wikipedia.org/w/index.php?title=Artificial_intelligence_in_fiction&oldid=1274274117