[ericy]

When different AI give different answers, something seems wrong.

 

[ericy]

These generative engines (like ChatGPT) are just predicting the next word based on text they’ve seen before. They don’t “know” facts and can’t do math, etc… use at your own risk! The technical explanation is “hallucination”… A lawyer used ChatGPT for legal filing. The chatbot cited nonexistent cases it just made up

The weird thing is that they can't even explain how AI hallucinations actually are happening.

 

[ericy]

I didn't. I don't even have ODBeleven. Just testing the approaches of interrogating Bard as posted above. If this works, I'll definitely buy an ODBeleven to test this out.

Ask it something g off the wall -like how use obdeleven to set the clock on the VCR.

 

[ericy]

I think you weren't expecting this response:


That said I can't reproduce OP's responses in chatGPT, only semi sorta in Bard.

I didn't know what to expect. I sort of half suspect that it was just making stuff up in some of the other responses. Maybe it was capturing things that people had tried along the way - hard to say. I suppose it realized that the question about the VCR was so far afield that it could reject it out of hand.

I would want to verify by closely looking at the obdeleven site/app before I bothered to try any of this with my own car.

 

[ericy]

I started looking at some VCDS scan logs, and my car doesn't even have modules 55 or 4B. Module 15 is airbags, not comfort. Module 13 is front radar, not roof. So it seems like some of the answers are useless hallucinations. So much for AI.

I have been doing reading on AI - honestly it is only as good as the training you give it, and how smart the rewards are. And it can find unexpected "solutions" that a human would never think of.