ChatGPT trips over easy poker questions

[Tero]

I wanted to see if I could design simple quizzes for myself with ChatGPT to drill some combinatorics practice.
But the poor thing started tripping from the start...








And this was not the only question that made it fail.

The takeaway from this - If you are not sure about something in poker and you want to ask from an AI, you can't be certain that you'll receive an accurate answer.

 

[Andyreas]

Yeah I had the same situation when I asked it a few poker questions months ago.

Well actually it's good. We're probably still ahead of AI for now. 🤪🤣

 

[Debi]

It is definitely not reliable for things like this. I asked it for the date I will reach full retirement for social security and it was off by 2 months. When I told it that it was incorrect it doubled down lol.

 

[louuu19]

ChatGPT is not reliable. I asked it about a matrix, and it got it wrong! So don't rely on it

 

[pentazepam]

The Chat-GPT that is free (or cheap) for the public is a LLM (Large LANGUAGE Model).

The math answers are by definition worse than its language capabilities.

But they are getting better and better faster and faster (exponentially).

If you want a specialized Chat-AI you can fine-tune even the smaller ones to be good (or at least better) at math.

Fine Tuning Gemma-2b to Solve Math Problems

Mathematical word problem-solving has long been recognized as a complex task for small language models (SLMs). To reach a good level of…

medium.com

[2310.10047] Improving Large Language Model Fine-tuning for Solving Math Problems

arxiv.org

 

[Tammy]

Chat GPT is hopeless with anything to do with numbers or dates, even the most simple of asks. I have gone round and round with it, and when I pointed out its errors - like Debi said it doubled down (after apologizing for being wrong lol).