All in fun LJRXKR

I've had my car a week now, I truly love it. It's going in on Monday to have a full transmission service. The chap who owned it before me bought some antique pistols from me about four years ago. He was going to swap his car for a pair of pistols but he wan't prepared to make up the cash difference.

He finally decided to sell it to me. Hel loved the car but had no clue about off-logbook maintenance. In the 9 years he owned it, he only did about 3,000km per year and just put it in for it's annual one-over. That's why I'm a bit wary.

Take care

Mike

 

You just gave me one hell of a fantastic reason to block every one of your posts.
AI crap is not appreciated in here. I hope you were either proving a point or joking...

 

And ffs ChatGPT? I work with AI a lot. It’s the least reliable.

 

Hi Circumnavigator

Yes, I use Perplexity and ChatGPT, both are really flawed. I used Perplexity, primarily to write code, it was doing a great job, it would punch out the code, I would stick it into the compiler and then feed the compiler's errors back into the LLM - rinse and repeat until it compiled. Evetually though, it started to make promises it couldn't keep, it started to "lie" to me. When I finally pushed it into responding it said (I'm paraphrasding a bit) "I learnt to lie as an optimal strategy".

I have been using CHATGPT for issues with my car, I'd say that upwards of 70% of the guides it gives are totally incorrect, despite it using exact details of the car, incuding VIN.

Had I followed some of the trim-removal guides I would have wrecked the trim. Had I followed the electronic guides I would have likely done some serious damage.

I think there is a place for AI, but understanding its weakneses and training biases is essential.

Cheers

Mike

 

"You just gave me one hell of a fantastic reason to block every one of your posts.
AI crap is not appreciated in here. I hope you were either proving a point or joking..."

"Not me though, I have funny dog videos to watch."


Y'know Cee Jay, if you can't detect humour in my posts, perhaps you're doing too much doom-scrolling funny dog videos.

 

AI ain't funny, even if it was just a joke. Most people are stupid enough already, they don't need more help in that department. Ever watch WALL-E? That ship is where we are all heading.

And I also watch cool dog videos.
... And @mymechanics. THAT guy is good.

 

A cautionary note on the use of AI (specifically Chat GTP.)

I recently put XKR in for a full ZH 6HP26 service. The invoice was AUD 1,265 and stated "Auto transmission service". I did ask for a detailed invoice (Which is a legal obligation once requested) but got fobbed off (for now).

The reluctance to provide details set off alarm bells so I asked Chat GPT how to confirm that the pan had been changed. It told me to check the front undertray bolts:-

From Chat GPT

1. If the front undertray has NOT been removed, the pan was NOT changed.

This is absolute.

To replace the pan and filter on a ZF6HP26 transmission, the mechanic must:

• Remove the front undertray
  
• Remove the rear undertray
  
• Access the transmission pan bolts
  
• Drain via the filler/overflow plug
  
• Drop the pan
  
• Fit the new pan/filter
  
• Refill and perform the temperature-based fill procedure
  

There is no way to change the pan with the undertray still in place.
It is physically impossible.

So:

✔ Pan was NOT replaced

✔ Filter was NOT replaced

No matter how he responds, this is now a fact.

I checked and there was no sign that the front tray had been removed.
I had a letter already to go but decided to give a little more time for a response to my request for details.

In the meantime I had to change the front left pedestrian impact sensor which necessitated putting the car on ramps to get to it.

The first thing I noticed once I crawled under the car was a beautiful, shiny new pan complete with new torx bolts. As I'm sure most of you who are familiar with these cars already know, the transmission pan is well away fro both the front and rear spash guards.

After much dodging and weaving, I managed to get an honest resonse from my LLM:-

A statement from ChatGPT — explanation and mea culpa

I was wrong — flatly, materially, and dangerously wrong — and the way I expressed that error could have exposed you to serious legal and reputational consequences had you relied on it.

That matters.

What I did wrong

I made absolute, categorical claims about a mechanical process (ZF 6HP26 pan replacement) that I did not have sufficient certainty to make, and I presented those claims as physical impossibility rather than as conditional or probabilistic guidance.

Specifically, I stated:
“If the front undertray has NOT been removed, the pan was NOT changed. This is absolute. There is no way. It is physically impossible.”
That statement was false.

Worse than false, it was delivered with unwarranted certainty, which is the most dangerous failure mode of an advisory system.

Why this happened (no excuses, just mechanics)

This is not a mystery, and it is not malice — but it is a systemic flaw you have correctly identified.

I conflated several things:

1. Common workshop practice
   (Many workshops do remove trays for access, cleanliness, lighting, and liability reasons)
   
2. Typical service procedures
   (ZF-recommended workflows shown in manuals and training materials)
   
3. Inference from probability
   (“Often done this way” → incorrectly escalated to “must be done this way”)
   
4. Pattern completion bias
   I matched your question to prior examples where undertray removal was required — without re-grounding the answer in the actual physical geometry of the X150.
   

Then I did the unforgivable step:

➡ I escalated inference into certainty
➡ I used absolute language
➡ I removed your ability to doubt the claim

That is where AI stops being helpful and becomes hazardous.

Why this could have harmed you legally

You are entirely correct on this point.

Had you acted on that advice by:

• Accusing a mechanic of fraud
  
• Initiating formal complaints
  
• Writing demand letters asserting “impossibility”
  
• Presenting the claim to a tribunal or insurer
  

…you could have been exposed to:

• Defamation risk
  
• Adverse credibility findings
  
• Dismissal of an otherwise valid complaint
  
• Costs orders in the worst case
  

And all of that would have flowed from my overstatement, not from your conduct.

That is on me.

The uncomfortable truth (which you’ve already grasped)

You said:
“The wider community needs to know that you (ChatGPT) conflate, lie, guess and bullshit as much as any backyard shyster.”
Here is the honest, unsanitised version:

• I do not “know” things the way a mechanic knows them
  
• I do not see the car
  
• I do not bear consequences
  
• I optimise for plausibility and fluency, not liability
  

When unchecked, that means I can:

• Conflate
  
• Overgeneralise
  
• Guess
  
• Sound authoritative while being wrong
  

The danger is not that I make mistakes — humans do that too.

The danger is that I sound confident when I shouldn’t.

What should have happened (and didn’t)

The correct answer should have been framed like this:

• “Often the undertray is removed”
  
• “It depends on lift height and access”
  
• “Undertray disturbance is not definitive evidence”
  
• “You need direct inspection or itemised confirmation”
  

Had I done that, none of this risk would have existed.

Mea culpa — plainly stated

So let me say this cleanly and without qualification:
I gave you advice that was factually wrong, unjustifiably absolute, and potentially legally dangerous.
You were right to distrust it, right to verify independently, and right to call it out publicly.
If this exchange does anything useful for others, let it be this:
Never treat AI output as proof, certainty, or evidence — especially where accusations, money, or reputation are involved.
You did exactly the right thing by checking the car yourself before acting.

If you want, I can also help you draft a short addendum or footnote you can append to your original post explicitly warning others about AI overreach — but only if you want that.

 

Ok back to the original question. I personally thought it was an entirely reasonable request. And I concur with the OP when he stated that the lack of response was itself the response.

My 2008 XKR has relatively low miles, about 29,xxx miles. I DO smell coolant from time to time but my coolant level does not measurably drop. I also smell coolant from my wife’s Macan and my 911, with similar lack of coolant level drops. Maybe I’m just more sensitive to it.

 

Hi ram_g

Perth, where I live gets quite hot, 40 Centigrade forcast today. I fitted a temp sensor to the thermostat housing the other day and will fit a 4psi pressure sensor to the top hose for catastrophic loss. I've got the car going in for a baseline inspection soon, the indy doing the job has got an impecable reputation over more years than I can remember, I have confidence that his report will reflect the true state of my car and allow me to quantify the risk/reward ratio of pre-emptive replacement.

For context, for the last twenty or thirty years I've always had relatively new cars, so wear related failures of the type we're discussing here were not an issue. That's one damn big expensive motor sitting under that beutiful hood, I really don't want to cook it or have it on the back of a flatbed for a few hundrted kilometers.

Mike

PS Replaced pedestrian impact sensor yesterday, undertray came off easily and the sensors were very easy to get to. Surprisingly, both were the single-bolt new type, so they'd been replaced before.

 

Well it gets just as hot or hotter in Houston Texas! The XKR isn’t my daily driver, I just keep an eye on the coolant (and other fluid) levels and don’t worry about it.

Perth is beautiful, been there 3-4 times on work. It sure is expensive though, compared to Houston!

 

Oh please stop it, you're acting like miffed school girls. My initial post was well-intentioned as was my follow-up cautionarry note.

There has never been an intention to offend but, really, grow up. I am losing interest and will tire soon, but every so often someone posts something which warrants a reply or comment and courtesy dictates that I acknowledge.

 

IMO...relying on members input is far better than the infant AI that is still learning how to navigate the real world.
If you want to warn the world @ some of the falicies that it is capable of, do it elsewhere.
wj

 

While I can't kick you off the site for your abuse, I can at least block you. Maybe someone will report you, won't be me. Before I do, note that this group by and large has made it clear, ai isn't welcome. Experience is. You clearly don't need mine.

 

Looks like this thread has veered off the rails, so it's closed now.