I DOVE INTO WORKING WITH DEEPSEEK

...And I’m Gobsmacked

Back in the day, there used to be a trope about “Your Intrepid Reporter.” That’s how I felt today as I dove into deepseek to see how it measures up as a writing partner.

The TL;DR

I’m blown away by the way deepseek “thinks out loud.” Although I’d read and heard about how deepseek “Shares its reasoning,” I was NOT prepared to see it think before answering: 

Okay, so I need to summarize the "big ideas" from this story document. Let me start by reading through everything carefully. 

Or 

First, I need to break down the audience segments more precisely. The core profile includes software engineers, managers, tech fans, and savvy users. But how do these groups consume media? For example, software engineers might frequent Reddit or Hacker News, while managers might prefer podcasts or LinkedIn articles. Tech fans could be on platforms like Twitter or follow specific YouTube channels.

In all of my previous posts, I placed a VERY high value on how AI responded to various elements in my story. My AI tools might be wrong, but I always made note of what it decided to respond to. 

Deepseek is a unique writing partner, in that it literally itemizes what is resonating in your writing -- at least what resonates with this LLM. 

SUBSCRIBE NOW

This might be the first post of mine you’ve read. My substack, “The AI Writers’ Room” is all about incorporating AI into your writers’ process. If you’re into that, subscribe!

And now back to deepseek:

SIGNING UP / STAYING CONNECTED

Here’s the worst I can say about deepseek: It is hard to sign up, and it’s so busy that it will often just not answer your question. Think “AOL BUSY SIGNAL” frustration. 

Second biggest problem: I don’t trust that it will remember our chats. I can’t set up projects. So, for now, it’s only good for one-offs. 

I gave deepseek the same test I gave Claude, ChatGPT and NotebookLM. Deepseek gave me something none of the others did -- it thought out loud. 

When I asked deepseek to summarize the big ideas in my story, it shared its thinking by listing the key points and themes, various things I’d mentioned in my notes as “central questions,” etc. And then it said:

Hmm, I'm not entirely sure how all these elements tie together. ... I think I have a rough idea, but to make sure, maybe I should ask some clarifying questions. 

WHOA! Now we’re cookin’. 

The summary of big ideas did not differ greatly from what Claude and NLM said. But then, without prompting, it suggested that I write for a streaming series, an interactive story, or a serialized podcast. It kind of anticipated what I was about to ask. 

MAKING THE STORY MORE POWERFUL

In addition to outlining audiences and where I can find them, deepseek really did a great job of digging in to how I might improve my story. Before it even asked me 5 questions, it suggested:

• Raising the stakes. (Okay, that’s standard development stuff.)

• Satirizing global locations to “regionalize” the story. I hadn’t really thought about that, although my notes clearly suggest that direction.

• Structuring for “bingeability.” Again -- I didn’t say anywhere in this summary that I was thinking about a streaming series. But it’s really beginning to make sense!

• “Anchoring” my absurdist humor. Good note. 

• “Hopeful resolution with ‘bite’” ... Which says to me “Deepseek, you get me!”

THAT’S ALL FOR NOW

We’re living in “Jevons Paradox.” (The more efficient a technology becomes, the more demand it will create.)  I have been trying to ask my fourth question for an hour now, and am going to give up for a while. 

Because it’s overloaded (I guess it DOES need some of those big expensive chips) and because I don’t trust it to remember me, I’m going to use deepseek sparingly. BUT... in full force, it would be a real threat to Claude. 

YOUR TURN

Let me know -- are you trying deepseek? What’s your experience?

And if you haven’t yet… please subscribe:

If you think you know someone who’d enjoy diving into the conversation, share this newsletter with them.