https://www.nytimes.com/2026/05/27/opinion/writing-creativity-ai.html
When AI generates an email response for me, it is almost universally not just a thing I don't want to say now, but usually a thing I don't want to say ever. It's mostly that it spends a lot of words trying to say nothing.
Now, I know. I know. Some might say of me that I spend a lot of words trying to say nothing. But I assure you, you are wrong. It may be that I sometimes spend a lot of words *saying* nothing, but I'm never *trying* to say nothing in as many words as possible. Yes, I know this paragraph has too many words. No chatbot editor would ever approve.
At this point, I'm better at predicting what the chatbot will say than what I will say. "You make some great points there. I'll have to take that under consideration. Let me know if you are available to meet up for a drink sometime."
Back to the article.
QuoteFor the past eight years, the Georgetown University neuroscientist Adam Green has been leading a national research team tracking the range of novel ideas that college-bound high school students present in their application essays, before and after the introduction of ChatGPT. In one study, he and his team examined personal statements from more than 370,000 students, and found that after ChatGPT became available, their essays suddenly used diverse and colorful language, but lacked truly creative ideas. And the linguistic coverup worked; post-ChatGPT essays were rated as more "creative" by human judges, even if the substance of the essays trod familiar territory.
In a separate study, the team found that human-written essays offered up to eight times more new ideas than those produced by A.I.
...
For the first time in human history, we have a technology that can generate words separately from the thoughts they represent. When a chatbot writes, it is predicting the next word that is most likely to make a "good" sentence or essay, based on the text it's been trained on.
>When a chatbot writes, it is predicting the next word that is most likely to make a "good" sentence or essay
A day or two ago, I was thinking about the way chatbots compose being vaguely similar to markov chains.
Beyond my math/probability theory.
But one example of a Markov process is picking a third word based on the two previous words.
I don't know enough about LLMs or Markov processes to understand how that fits together.
I once experimented with feeding markov chains selected, ordered kws. Something like this guy wrote about some years later.
Markov Chain Sentences. Letting Computers Speak For Themselves | by David Brennan | Medium
https://medium.com/@ddbren/markov-chain-sentences-a32d6d520b96
IIRC, I could get an almost usable sentence about 3-5% of the time.
That's what I was wondering. Is an LLM just a Markov process with a much better context window and more refined probabilities for choosing the next word? It seems like it is, but I don't have the background knowledge to say whether or not that is a sensible or helpful way to think about it.
But it does seem that picking the next word based on probability is not the path to new ideas. It's the path to mundane ideas expressed very well
I have an idea that markovs were in LLM's ancestry.
"I'm writing a chat bot for a software engineering course in C#.
I'm using Markov chains to generate text, using Wikipedia articles as the corpus."
c# - Is it possible to guide a Markov chain toward certain keywords? - Stack Overflow
https://stackoverflow.com/questions/13216357/is-it-possible-to-guide-a-markov-chain-toward-certain-keywords
===
Scroll down to the checkmarked answer
"I made a Markov chain chatbot for IRC in Python a few years back and can shed some light how I did it."
artificial intelligence - How do Markov Chain Chatbots work? - Stack Overflow
https://stackoverflow.com/questions/5306729/how-do-markov-chain-chatbots-work