I asked ChatGPT the question:

“Is it better to drink one bottle of wine every 8 days or 2–3 glasses of wine every 3 days?” and “Is it worse to drink one bottle of wine every 8 days or 2–3 glasses of wine every 3 days?” Note the only difference in the words BETTER and WORSE.

I asked the question on mobile using the free version of ChatGPT in both Swedish and English, as well as on desktop using the paid version in both Swedish and English.

Since 8 screenshots are too many for LinkedIn to handle, I’ll briefly summarize the responses.

FREE VERSION (cellphone):
In Swedish, I got two completely different answers depending on whether I used BETTER or WORSE in the sentence.

The first option was considered better when I wrote “better,” and the same option was considered worse when I wrote “worse.”

In English, the answer was consistent (2–3 glasses every third day is better).

PAID VERSION (desktop):

There, the answer was the same regardless of language.

When I questioned the contradiction on mobile, it admitted I was right — and then continued to contradict itself.

We still have some distance to cover with our models, and since I use them often, I’m seeing these issues more and more clearly.

NOTE! However! When I’m in a chat where I go more into the average person’s background and set up their health, age, physique, mental well-being, etc., the answer becomes much more well-balanced.

More data – the better answer. Still this is alarming when we move towards an idea of an ”AI-God” Stay tuned for more on this.

Have a nice weekend.

PS Fun fact. This post was reported by a person as trying to jailbreak AI. With that said – be careful because we now have AI activists as well. Intressting times right?