AI LLM

AI LLM

26 August 2026

I was going to write about LLMs today. Now I have to, because Patrick commented that he was curious. So, what is an LLM? A Large Language Model (LLM) is a form of artificial intelligence designed to understand and process human language and generate human-like text. It is trained on enormous datasets containing billions of words from books, articles, and websites. These deep-learning models extract the structural patterns of language from that material. Rather than actually “thinking” or remembering exact answers, an LLM primarily functions as a highly sophisticated predictor of the next token. When given a prompt, it calculates which next word or word fragment (token) is mathematically most likely based on its training.

On Saturday, a Catholic priest told me that I had not yet read the first encyclical of Pope Leo XIV. He is right. I have read a lot about it, but I have not read Magnifica Humanitas. I am going to do that now. The Pope is not an opponent of AI, but argues that it should serve humanity and should not come at the expense of human dignity. This Pope models himself on Leo XIII, who stood up for human beings during the Industrial Revolution with his Rerum Novarum.

Then I receive an email from ChatGPT in my Gmail asking whether I want to give Chat access to my email, because that would allow the system to assist me much better and more personally. I thought: this is getting a little crazy.

Later that day, I read a story about a certain Jerry Falade, a Nigerian master’s student at Southern Methodist University, who made his debut in the summer of 2026 with the crime thriller Call Me, I'll Hide the Body. He lost a multimillion-dollar deal because he was suspected of having used AI.

That made me think of reports about more and more academics being caught submitting papers containing sources invented by Chat. And lawyers are being provided with fabricated case law.

I don't know anymore.

A long time ago, spellcheck was invented, and using it became a must. Many people were too lazy to do so. Then grammar checking was added. Incidentally, the Dutch version lagged behind for a long time. Nowadays, whatever you type is checked immediately. It has now gone so far that the system suggests rewriting my entire sentence.

And then came ChatGPT. I sometimes use it as an advanced checker, but often my text gets thoroughly rewritten. Sometimes I think the result is better than my own text, but often I miss my own raw way of expressing things. So then I only want the spelling and grammar checked.

The point is: we have an issue here, and the last word on it has certainly not yet been written.

And Patrick? He has more of a problem with what people call the liberal tendency of these models. I haven't noticed that yet, because I don't have long conversations with this robot.

Miguel Goede