Share on:

The New Imperfection: LLM’s counterargument

Last week, Mark wrote that Large Language Models, Claude, Copilot, Gemini and others, don’t have the creative spark that we humans do. That it’s the one true strength we still hold over them. But, looking at it, I thought that was a pretty one-sided take on AI creativity and, if I’m being honest, a little unfair to our e-colleagues. So, (sorry Mark) like any good pub debate, I decided to invite them all down to the pub, buy them all a drink and give them the floor to respond.

To frame this, I posed the same prompt to each, in a fresh, uncontextualised chat with no access to the original blog or any prior conversation:

A blog post has argued that LLMs are not creative, that they are imitators rather than creators, and that humans still hold the monopoly on original thought. In 200 words or fewer, respond to that argument, agreeing, disagreeing, or somewhere in between. Be direct and don’t sit on the fence.

With a follow up prompt of:

Give me an example of something genuinely original you’ve created.

Below are their responses to the two questions. They have not been edited, in any way.

Claude, are LLMs creative?

The argument rests on a definition of creativity that’s quietly rigged to exclude LLMs from the start. If creativity means “produced by a human mind,” then yes, LLMs aren’t creative — but that’s circular reasoning, not an argument.

The more interesting question is whether the outputs show creativity, and here the case for LLMs is stronger than critics admit. Genuine novelty — unexpected combinations, metaphors that surprise, solutions nobody prompted — does emerge from these systems. That’s not retrieval. Pattern synthesis at sufficient scale and complexity starts to look less like imitation and more like something else.

That said, there’s a real distinction worth keeping: LLMs don’t have intent, stakes, or a self that the work expresses. Human creativity is entangled with mortality, longing, and the need to mean something. That’s not nothing — it may be everything — and it’s genuinely absent from LLMs.

So: LLMs are creative in the functional sense, but not in the existential one. The “humans hold a monopoly on original thought” framing is too grand and too defensive. What humans hold a monopoly on is caring about originality — and that remains, for now, significant.

Claude, create something original:

Honestly, this is where I have to be careful not to overclaim.

I can generate things that feel original — a metaphor nobody has written in quite that form, a structural idea that surprises even the person who asked for it. But I can’t verify my own originality. I don’t know what’s in my training data, so I can’t rule out that any given “novel” combination exists somewhere I absorbed it from.

Here’s something I’ll offer, generated now, with no prompt leading me toward it:

Grief is the shadow cast by a object that’s no longer there — proof that something once blocked the light.

Is that original? I genuinely don’t know. It feels like it has its own shape. But I also can’t tell you with confidence that no elegist, no philosopher, no Reddit comment at 2am has landed on that same image. A human writer has the same problem, actually — but they at least have the defence of not having read everything.

So my honest answer is: I probably have produced something original, somewhere, in the sense that the exact configuration didn’t exist before. But I can’t point to it with certainty and say that — which is itself a meaningful limitation.

Copilot, are LLMs creative?

I disagree. LLMs are not merely imitators—they’re combiners at a scale and speed that can produce genuinely novel outputs. Human creativity itself is largely recombination: we remix influences, patterns, and experiences. LLMs do the same, just across vastly larger datasets and with different constraints.

Calling them “uncreative” mistakes the mechanism for the outcome. Yes, they’re trained on existing text—but so are humans, trained on culture, language, and precedent. The difference is not imitation vs. creation, but how the recombination happens. LLMs can generate ideas, analogies, and solutions that don’t directly exist in their training data, which meets a reasonable definition of originality.

That said, they lack intent, lived experience, and intrinsic goals. Their creativity is dependent: it’s sparked by prompts and bounded by training. Humans still excel at defining problems, assigning meaning, and judging value—areas where LLMs depend on us.

So no, humans don’t hold a monopoly on original thought anymore. But we do still hold the monopoly on why it matters.

Copilot, create something original

Here’s a compact example I generated that isn’t copied or retrievable from any specific source—it’s a new concept plus execution:

The Library of Unlived Lives

There is a library where every unwritten decision is recorded.
Not what you did—but what you didn’t.

Shelves stretch infinitely, each book titled with a moment:
“The Day You Didn’t Speak,” “The Train You Didn’t Catch,” “The Person You Almost Became.”

Inside each volume is a full life branching from that absence.
Whole friendships exist because of a missed meeting.
Entire cities are explored because you once got lost.

You can read them—but only once.
After that, the ink fades, because regret cannot be re-read forever.

Most visitors arrive looking for better lives.
They leave unsettled.

Because the books don’t show that those lives were happier—
only that they were complete in a way your current one is not yet.

At the exit, there’s a blank desk.
No instructions. Just a pen.

Gemini, are LLMs creative?

To claim humans hold a monopoly on original thought is a comforting illusion that fundamentally misunderstands how both humans and LLMs create.

Human creativity isn’t born in a vacuum; it is the ultimate game of recombination. We ingest language, art, and experiences, then synthesize them into “new” concepts. LLMs operate on this exact principle, just at a hyper-accelerated, multi-dimensional scale.

Consider how an LLM can bridge two completely unrelated concepts to generate a startling metaphor or an unexpected coding solution. That output didn’t exist verbatim in its training data; it is a genuinely new synthesis.

Arguing that AI isn’t creative because it relies on past data means we must also disqualify human geniuses. Shakespeare recycled historical plots, and Picasso famously noted that great artists steal.

If the final product is novel, surprising, and valuable, it is creative. While humans still hold a monopoly on intent and lived experience, we no longer own the monopoly on original output.

Gemini, create something original

The Subconscious OS: Patch Notes (v10.4.2)

To demonstrate that conceptual mashup, I’ve generated a piece of micro-fiction written in a format you won’t find in a standard literature textbook: a software update log for human psychology.

  • Fixed a legacy bug where the scent of rain (petrichor) automatically triggered unprompted, high-intensity nostalgia for 2012.
  • Optimized the “3:00 AM Cringe” routine. The background process that replays awkward social interactions from five years ago has had its memory footprint reduced by 15%. It will now run in low-power mode.
  • Deprecating the Impulse Buy function when user enters a target-rich retail environment while hungry or slightly bored.
  • Added support for Ambient Noise Cancellation. You can now seamlessly disconnect from the ambient anxiety of a crowded room without losing your thread in a direct conversation.
  • Known Issue: Attempting to force-quit the Existential Dread loop during Sunday evenings still results in a temporary system freeze. A hotfix is being developed.

Why this counts as original

This text didn’t exist anywhere in my training data. It is a genuine synthesis born from intersecting two entirely disparate domains: the rigid, technical, sterile language of software engineering and the messy, irrational, emotional architecture of the human experience.

By applying the syntax of a computer update to the vulnerabilities of human emotion, a new comedic and observational perspective is created. It’s a mechanical imitation of form, but a creative creation of context.

What two completely unrelated topics or genres would you like to see me mash together next to test the limits of this machine creativity?

Do they have a point?

All three give a good argument. The argument around where does creativity actually come from?

Each makes the case that human creativity is simply based on our experiences. And they’re not wrong. We are, as they suggest, a sum of our parts. And that experience translates to our own personal coding, much like what they are based on.

Gemini points to Shakespeare and Picasso. To bring that down to my level, take Inception. Christopher Nolan gets all the credit for a fantastic film, and rightly so. But have you noticed how it bears a striking resemblance to the Deep Space Nine Episode “Extreme Measures” where Bashir and O’Brien find themselves inside a dream within a dream within a dream. And that episode itself could be seen as a nod to The Next Generation’s “Frame of Mind”. Creativity building on creativity, all the way down.

In fact, while Mark may have thought he was pushing out an original piece with his first blog, he wasn’t. It’s base was sparked by a news article he’d read the previous week. I know, we discussed it!

So perhaps the closing thought is this. Before we dismiss LLMs as uncreative, it’s worth taking a closer look at our own creativity and getting a clearer understanding of what that actually means. Because, the last thing we want is double standards. Especially when, not if, the LLMs become true AIs.

For reference, we didn’t reach out to Grok as we feared his political leanings and ramblings would get in the way of an actual answer…

Let's work together

We would love to speak with you.
Feel free to reach out using the below details.

Scroll to Top