The Quiet Crisis of Original Thought in the Age of LLMs

There has probably never been an easier time to produce something that looks finished. A proposal, an essay, or a dozen plausible ideas can now appear in seconds. Large language models have dramatically lowered the cost of expression. That is an extraordinary achievement. It also creates a less obvious problem: when producing a competent answer becomes cheap, it becomes increasingly easy to confuse competence with thought.

The central question of the LLM age is whether humans will continue to think.

When good enough becomes almost free

For most of history, producing polished work required effort even after the thinking was done. You had to find the words, organize the argument, and wrestle an idea into form. That friction was inconvenient, but it also forced the creator to spend time with the idea.

LLMs remove much of that friction. This is enormously useful when the thinking already exists. A person with a clear argument can use an LLM to structure it, challenge it, compress it or express it more effectively. But the same tool can also allow someone to bypass the formation of the argument altogether. Instead of asking, “What do I believe?”, it becomes tempting to ask, “What should I say?”

Those are not the same question.

The first requires judgment. The second can be outsourced.

The strange diminishing return on originality

There is a paradox emerging around creative work. At precisely the moment when original thinking should become more valuable, the immediate reward for doing it can feel smaller.

Imagine two creators. One spends days reading, observing, arguing with an idea, discarding obvious conclusions and eventually producing a perspective that could only have come from that particular mind. Another produces twenty plausible variations with an LLM in an afternoon. In many environments, both may initially receive the same reward: a post gets published.

From the creator’s point of view, this can make originality look economically irrational. Why spend ten hours developing one unusual idea when ten acceptable ideas can be generated in ten minutes?

There is pressure from the other direction as well. Audiences are becoming accustomed to the smoothness of machine-assisted communication: tidy frameworks, balanced paragraphs, familiar hooks, predictable visual aesthetics and the reassuring cadence of something that sounds approximately like everything else. As more of this material enters the culture, familiarity itself can begin to masquerade as quality.

The danger is not that everyone will suddenly prefer bad work. The danger is subtler: our definition of good work may gradually collapse toward whatever is easiest to recognize.

LLMs have a natural relationship with the centre

A language model is exceptionally good at navigating patterns that already exist. It has learned from vast amounts of human expression and can recombine those patterns with remarkable fluency. That makes it powerful precisely because it has access to so much of what people have already said.

But originality often begins somewhere less comfortable: with an observation that has not yet become conventional, a contradiction nobody has resolved, a preference that cannot yet be justified by consensus, or a question that initially sounds slightly wrong.

The machine can help explore that territory. It can interrogate an argument, offer counterexamples, expose assumptions and connect distant ideas. But somebody still has to decide which direction is worth walking in.

That decision is taste.

And taste cannot merely mean choosing the answer that looks most like existing good answers. Taste is the capacity to notice the difference between what is technically impressive and what is actually worth saying.

Critical thinking becomes an act of authorship

Critical thinking is often described as the ability to evaluate information. In the age of generative AI, that definition is no longer sufficient. Critical thinking also means maintaining ownership of the questions themselves.

Why am I asking this? What assumption am I accepting without noticing? Is this conclusion true, or merely well phrased? What would I believe if nobody had shown me the conventional answer? Which part of this argument comes from evidence, which part comes from fashion, and which part comes from me?

These questions matter because an LLM can provide an extraordinarily convincing first answer. And convincing first answers are dangerous when they eliminate the psychological need for second thoughts.

The skill we need is not resistance to AI. It is intellectual sovereignty while using AI.

A strong thinker should be able to use a model aggressively without becoming intellectually dependent on it: ask it for twenty possibilities and reject nineteen; ask it to attack a cherished idea; ask it what the conventional position is so that one can examine whether the convention deserves to survive. The machine becomes a sparring partner rather than an oracle.

Originality is not novelty for novelty’s sake

There is an equally unhelpful response to all of this: trying desperately to be different. Originality is not eccentricity. It does not require disagreeing with consensus simply because consensus exists. Sometimes the obvious answer is correct.

Original work comes from reaching a conclusion honestly rather than selecting one for its familiarity. Two people can independently arrive at the same conclusion and both have thought originally. Conversely, someone can produce an unusual-sounding idea that is intellectually empty.

The distinction is provenance. Where did the idea come from? Was it inherited, generated, imitated or examined? Can the person behind it defend it, modify it and explain what evidence would make them abandon it?

That is why thinking for oneself matters more than simply sounding different.

What happens when the audience changes too

Creators are only half of the equation. Culture is shaped by what audiences reward.

If people consume enough machine-generated and machine-assisted material, certain patterns can become increasingly familiar: the same pacing, the same vocabulary, the same structures, the same visual references, the same emotional beats. Eventually the unusual work can feel wrong not because it is worse, but because it violates an increasingly standardized expectation of what polished work is supposed to look like.

This is how taste can narrow without anyone consciously choosing to narrow it.

A public accustomed to frictionless competence may become less tolerant of the rough edges that sometimes accompany genuine discovery. Yet some of the most interesting ideas arrive unfinished. New artistic languages initially look strange. New arguments often require effort. New categories rarely resemble the best existing example of an old category.

If our collective taste becomes too optimized for immediate legibility, we risk rewarding the statistically familiar over the intellectually alive.

The premium will move from production to perspective

This does not mean originality will disappear. It means its economics will change.

When everyone can produce a respectable paragraph, respectable paragraphs stop being a meaningful advantage. When everyone can generate attractive concepts, generating concepts becomes less scarce. When everyone can imitate the surface characteristics of expertise, the surface characteristics of expertise become less useful signals.

Scarcity moves upstream.

The valuable questions become: Who noticed something first? Who framed the problem differently? Who developed judgment through years of attention? Who has lived experience the model cannot substitute for? Who is willing to hold an unpopular idea long enough to test it? Who can distinguish an elegant explanation from a true one?

The competitive advantage of the future may therefore look less like superior production and more like superior perception.

Use the machine. Keep the mind.

There is no virtue in performing manually what a machine can do better simply to prove that a human was involved. Writers should use LLMs. Editors should use them. Researchers, designers, entrepreneurs and publishers should use them. Refusing leverage is not originality.

The boundary worth protecting lies elsewhere.

We should outsource the expansion of possibilities without outsourcing the choice between them. We can outsource drafts without outsourcing convictions. We can ask machines to help us see more, but we should remain responsible for deciding what deserves our attention.

Because as generation becomes infinite, selection becomes identity.

What you choose to believe, reject, emphasize, pursue, publish and defend will increasingly be what distinguishes your work from a machine-generated average.

The courage to have a mind of your own

The deepest risk of the LLM era is therefore not that artificial intelligence becomes capable of thought. It is that humans become comfortable avoiding it.

Thinking is slower than generating. Developing taste is slower than prompting. Forming an original worldview is slower than asking for a summary of the existing ones. For a while, the market may not always reward that extra effort visibly or immediately.

But originality has never been valuable because it was efficient. It is valuable because civilization moves when somebody sees what everybody else has been looking at and notices something different.

In an age that can generate endless answers, perhaps the distinctly human advantage will belong to those stubborn enough to continue forming their own questions, and disciplined enough to decide for themselves which answers are worth believing.