LLMs (Large Language Models) have made a huge impact on how we work, communicate, and essentially live. There is already a lot of talk about safety, mental health, misinformation, and other mainstream issues, but not nearly as much about what I see as one of the inevitable consequences of all this: getting dumber.
Today, I want to talk about a behavioral problem.
Suppose someone has an idea. They want to work on it, develop it, and turn something that is currently abstract into a new reality.
That is pretty much the backstory of every human advancement ever made.
But what if part of that thinking process, or even the whole process, could suddenly be delegated? What if instead of having to organize our thoughts, question them, connect the missing pieces, and decide what we actually want, we simply provide a stream of unfinished thoughts, ideas, and half-baked goals?
Traditionally, that would constitute a project doomed to fail.
Not anymore.
With a bit of help from one’s favorite LLM, even the most unprepared ideas can be pushed to surprising lengths. ChatGPT, Gemini, Grok, Claude, and the others are getting better and better at “understanding” the wider picture and pinpointing the exact parts that matter, even when those parts are buried somewhere inside the chaos we communicated to them.
That is good for all of us, isn’t it?
Turns out that it may not be.
Yes, the work is done faster. Yes, decisions can be better grounded because we can process more information and consider more angles than before.
But what happens to our own ability to think and express ourselves?
It deteriorates.
The same way our muscles deteriorate if we do not use them, our ability to structure thoughts, connect ideas, formulate what we want, and communicate it clearly can deteriorate if something else constantly does that work for us.
And I think this goes beyond the individual.
Talking to people as if they were LLMs
One of the reasons LLMs are so useful is that they are incredibly forgiving.
I do not necessarily have to explain myself perfectly. I can start with one thought, jump to another, change direction halfway through, leave out context, contradict something I said earlier, and still expect the model to figure out what I am trying to achieve.
Quite often, it does.
That is one of the impressive things about modern LLMs. They are becoming increasingly good at reconstructing intent from messy input rather than requiring us to formulate that intent clearly in the first place.
The problem starts when we get used to communicating like that.
After spending enough time talking to systems that constantly fill in the blanks for us, it becomes very easy to communicate with another person in exactly the same way. We send thoughts as they arrive, leave connections unstated, jump between ideas, and assume that the other person will understand the wider picture.
Except the other person is not an LLM.
They are not a model specifically trained to interpret whatever mixture of context, assumptions, previous conversations, and half-finished thoughts we just sent them.
So what happens?
Increasingly, they use an LLM to decipher it.
Someone sends a poorly structured message because they have become accustomed to being understood anyway. The recipient takes that message, gives it to ChatGPT or another model, and asks what the person actually means or what they are supposed to do with it.
The communication technically worked.
But I am not sure that should make us feel better.
What about communication?
This is where I think the issue becomes much bigger than someone becoming slightly worse at writing an email or organizing an idea.
Humans are already a connected species. Almost everything we build professionally depends on our ability to transfer thoughts, intentions, knowledge, and decisions between one another.
If LLMs start changing the way individuals think, they will inevitably start changing the way those individuals communicate with each other.
At first, one person becomes less disciplined about expressing an idea because an LLM can understand them anyway. Then another person becomes accustomed to using an LLM to interpret unclear communication. Nothing necessarily breaks. The work still gets done.
That is exactly what makes this interesting.
We can lose part of the skill without immediately losing the result.
The LLM fills the gap.
And because the result still arrives, there is very little pressure to regain the ability that was lost.
Before LLMs, unclear communication created friction. If I could not explain what I wanted, another person would misunderstand me, ask questions, or force me to think about it again. If my thoughts contradicted each other, that contradiction had a much greater chance of becoming obvious.
That friction was annoying, but it was also useful.
Now there is an increasingly capable system sitting between the messy thought and the useful result, quietly compensating for whatever was missing.
The danger is that we continue functioning perfectly well while becoming less capable of doing some of the things that made that functioning possible in the first place.
We can still produce the report. We can still make the decision. We can still send the message. We can still understand what someone probably wanted.
But more and more of the cognitive work required to get from one point to the other is happening somewhere else.
That is why I think “LLMs are making us dumber” is not as dramatic a statement as it may initially sound.
If a tool continuously compensates for a skill that we stop exercising, there is little reason to expect that skill to remain equally strong.
And when that skill is not just writing, but the ability to organize thoughts and communicate them to another human being, the consequences enter the communication between all of us.
A quick note
These are my own observations and thoughts, not a scientific conclusion.
Researchers are also looking at the cognitive effects of LLM use from a more scientific perspective, and one study I came across is Your Brain on ChatGPT by MIT researchers.
