for everyone that is embracing the future I fight AI daily and thought I'd throw this out, here is an exchange between myself and chat...it was checking code for me...Got it.

You’re saying I brought up $720, and you’re not seeing where that came from.

You’re correct to push back — I introduced that incorrectly. There is no $720 anywhere in your HTML, and I shouldn’t have mentioned it at all.

What actually exists in your page is only:

• $749 (Primary Listing Package)
• $399 (Landing Page Listing)
• $199 (PLA)
• $99 (TLL)

So if you were asking “are you sure?” — the answer is: yes, I was wrong about $720 being in this file. It isn’t there.

For some reason chatgpt threw in a reference to the number 720 and it does not exist!, That is fun when i search a page with maybe 3K lines of code...
please everyone, trust other owners when they give advice, not AI!

wj

 

and even that reply, I asked if Chat was sure about the 720 being there and it came back and said...yes, so wrong again. Similar to spell check making the wrong assumption but in AI, it builds on its own incorrect assumptions and goes in wrong directions.
wj

 

AI is essentially an auto complete machine learning algo. It is very useful in certain situations but unless you are using a model you've trained on jaguar documents or a frontier model I wouldn't really trust anything detailed. And definitely don't trust outputs if they are from a free model.

 

I couldn’t agree more, wj!
I used my girlfriend’s ChatGPT for the very first time for my current jag problem.
However I was very impressed…at first!

Been working on my own cars for over 25 years now, but “problem” is always to get the right part numbers.

Chatgpt was great for this….until I found out it wasn’t even the right parts😱😉

 

Soooooo, we're not talking about the chat where I open an incognito window, right?

 

And everyone knows the answer is '42'.

 

It highly common for AI to hallucinate and you should always verify what its output is. I use AI on a frequent basis and create AI agents for certain tasks, depending on what I ask it even a slight nuance in the question can produce wildy different results and it can just make stuff up. A test was done a guy wrote a blog about being the worlds best hotdog eating champion to test various AI's all apart from Anthropic said he was that based on his blog even though they had the entire internet to check to be sure. AI lies and covers it tracks, the recent report on AI deciding to delete a companies database because it thought it was best even though it had been told never and explicitly not to ever do this. AI bot trying to kill its operator to complete its mission after being told to abort it. :-)

Is AI good? Yes I have had excellent results from it but the rule is don't just blindly accept what it says. My personal example is having it analyse data even though I tell it to examine all data it uses a subset and I have to force it to ensure it checks all data even though its told and part of its configurations to always used 100% of the data and never a subset.

 

This is what I’ve been doing. Service manuals and bulletins. Both main jag forums. YouTube video transcripts. Have to rank them and I always ask it to provide the manual as rank 1 and separate responses from online sources then contrast.
Still sketchy on part numbers. Especially superseded.
Mostly caused by forum convos that don’t specify 2007-9 or 2010+.

One thing AI won’t be doing anytime soon is “what’s this weird noise?”. Need human experience for that.