Level 1 (4 points)

ChatGPT keeps changing the lingo on my Mac

I have a question, and I'm genuinely trying to understand what is happening here.

Why does ChatGPT sometimes change the lingo or seem to distort the topic I'm writing about? Or is it just me because I'm multilingual?

The strange thing is that I keep the same terminology throughout the document I'm writing. I'm not deliberately switching languages or changing the way I describe something. Yet sometimes ChatGPT seems to decide that what I've written must mean something slightly different. Quite helpful, really, when you're trying to write something and suddenly find yourself discussing a completely different subject.

Then there is the other problem: spellcheck isn't always correct either.

There are words in English that are also words in other languages, and sometimes they're spelled exactly the same—or close enough to cause confusion. Take “desert” and “desert.” You can see the problem immediately. Same letters, completely different meaning and pronunciation.

This is exactly the sort of thing I'm talking about. There are quite a few other examples as well, but this is probably the easiest one to explain without turning this post into a linguistic thesis.

So I'm genuinely interested in your views.

Is this something that happens because I'm using several languages? Is the software simply guessing which language I'm using from the surrounding text? Or am I somehow managing to confuse the poor thing?

I'm not looking for an argument here. I'm actually trying to understand why it happens and whether there is something I can do about it.

Thank you for reading my post.

[Re-Titled by Moderator]

Original Title: Slightly confused question!

Posted on Sep 20, 2026 2:17 AM

4 replies

Grant Bennet-Alder

Level 10 (148,617 points)

Sep 20, 2026 7:44 AM in response to Sweet_Cicero

AI does not KNOW anything.

AI trained on an enormous number of writings from a huge variety of sources. It did not train on the 'best' sources. Then tried to "guess better" by having its results scored, then fed back through the system.

Most of what AI produces is utter garbage, including hallucinations. It deserves the term "AI slop". Without you deciding what is "Best" it is a hopeless mess.

If using it as a first pass can make you somewhat more productive, that is great. But taking its first whack at something as the final result is a BIG mistake.

Alancito

Level 7 (31,566 points)

Sep 20, 2026 8:35 AM in response to Sweet_Cicero

Sweet_Cicero wrote: "...Why does ChatGPT sometimes change the lingo or seem to distort the topic I'm writing about?..."

When ChatGPT does that, have you asked it to explain itself? (It may be able to.)

õ¿õ¬

MrHoffman

Level 10 (169,253 points)

Sep 21, 2026 8:42 AM in response to Alancito

Alancito wrote:

Sweet_Cicero wrote: "...Why does ChatGPT sometimes change the lingo or seem to distort the topic I'm writing about?..."

When ChatGPT does that, have you asked it to explain itself? (It may be able to.)

Ha! Thanks for the best laugh I’ve had so far today; a reminder of an “explanation” from a large language model :

Machine learning is absolutely useful, even if we’re still too often not sure why something works, and even if that uncertainty produces output quite possibly to our own detriment.

Is statistically-generated word salad from a large language model useful? Not so much. Well, not if you need the results to be correct. It can require as much or more time verifying the results as the large language model may purportedly have saved. Or more. (See image above.)

And that’s before we discuss the collection of the corpora, and what a Microsoft exec has referred to as the “largest theft of labor in human history”. And lead to the content scrapers that are now leading to those “verify you’re a human” tests all over, including the human-verification check here at Apple.

And that’s before we discuss how large language model output can be tuned to be obsequious and sycophantic, too. But turn those output dials far enough, load the appropriate corpora, and you get easily-automated romance scams. Or you get a middle-management memo generator. At least until the middle-management memo generator gets you dragged into court, gives away your product or service, or otherwise screws up.

Yeah. It’s going to be a bumpy ride. Enjoy your desert. Or your dessert.

Sweet_Cicero Author

Level 1 (4 points)

Sep 20, 2026 7:49 AM in response to Grant Bennet-Alder

Thank you for replying to me. It was very helpful.

I did not know this.

Thank you again.

Originally posted at discussions.apple.com/thread/256361050