Using ChatGPT for diagnosis

[red/green hawk]

Chat GPT helped me to diagnose a problem with my Insane Audio head unit in which the radio quit working. Insane Audio is no longer a company and has no customer service available. So, ChatGPT helped me pinpoint the problem which was a radio tuner chip and was even able to help me take the head unit apart and how to diagnose which chip I needed and even how to remove it. So, I was happy with the outcome.

I have an occaisional clunking noise coming from underneath near the rear differenital when downshifting from 4-3, 3-2, and sometime 2-1. And also an occaisional clunk from the same spot when accelerating around a round-a-bout that I live by. The clunking started around 45,000 miles. I'm now at just over 50,000 miles. ChatGPT stated that it is more than likely that the issue is the grease on the rear driveshaft at the slip yoke has dried up and that it needs to be regreased. It also states that it isn't that big of a problem.

There is a TSB regarding this issue per ChatGPT. It also stated it could be just normal driveline lash which Jeep acknowledges as a common annoyance.

1) Do you think ChatGPT diagnosed this correctly?
2) How easy is to grease the rear driveshaft slip yoke? Have any of you done it?

Sponsored

 

[ericw.]

I doubt that grease on the rear driveshaft is the main culprit since your problem persists only while turning but that's an easy thing to fix if your driveshaft has a grease fitting.

AI is an awesome tool and IMO it's worth the 30 seconds to ask AI just to see what it says. The more info you feed the AI, the better it answers.

With that, the only AI that I take seriously is Gemini and ChatGPT. Any of the others like Copilot and Grok just aren't at the same level yet.

 

[red/green hawk]

I doubt that grease on the rear driveshaft is the main culprit since your problem persists only while turning but that's an easy thing to fix if your driveshaft has a grease fitting.

It also happens on slow downshifts from 4-3, 3-2, and 2-1. Which ChatGPT attributes to normal driveline lash and/or the grease being dried up. It also stated larger tires can exacerbate the issue. Which I have 34s.

 

[Sweetums]

Chat GPT helped me to diagnose a problem with my Insane Audio head unit in which the radio quit working. Insane Audio is no longer a company and has no customer service available. So, ChatGPT helped me pinpoint the problem which was a radio tuner chip and was even able to help me take the head unit apart and how to diagnose which chip I needed and even how to remove it. So, I was happy with the outcome.

I have an occaisional clunking noise coming from underneath near the rear differenital when downshifting from 4-3, 3-2, and sometime 2-1. And also an occaisional clunk from the same spot when accelerating around a round-a-bout that I live by. The clunking started around 45,000 miles. I'm now at just over 50,000 miles. ChatGPT stated that it is more than likely that the issue is the grease on the rear driveshaft at the slip yoke has dried up and that it needs to be regreased. It also states that it isn't that big of a problem.

There is a TSB regarding this issue per ChatGPT. It also stated it could be just normal driveline lash which Jeep acknowledges as a common annoyance.

1) Do you think ChatGPT diagnosed this correctly?
2) How easy is to grease the rear driveshaft slip yoke? Have any of you done it?

First of all, Chat GPT and other "AI" LLMs are nothing more than glorified text prediction. They use complicated math to predict what series of words are most likely to be used related to the prompt. The rush to AI-all-the-things is happening at the expense of reality, be very cautious about trusting AI for factual information.

Clunking under acceleration is very unlikely to be related to the slip yolk, this part telescopes in and out to allow the axle to move up and down. When the axle droops, it gets farther from the transfer case and thus the drive shaft has to get longer. When it compresses the distance between the transfer case and diff is shortened. Think of it as a right-triangle. The chassis is straight, and you measure straight down to the diff. The drive shaft is the hypotenuse, if the leg of the triangle that is the distance from the axle to the chassis is lengthened, they hypotenuse (drive shaft) also has to lengthen.

Since the slip yolk only slides in and out, there's not really any lash in this part of the system. Even if there was, grease won't fix lash, which is due to bad adjustment or wear. Clunking could be a number of things from bad bushings in the control arms to worn U-joints in the drive shaft, or excessive wear or poor adjustment inside the differential. It could also be worn out engine mounts, as the engine's torque pushes it around in the engine bay, but that's unlikely given the young age of our trucks.

Short answer, I'm 99% sure Chat GPT is full of shit and regurgitating some nonsense it cobbled together from clueless forum and Reddit posts.

 

[SaiintNick]

I hate the world we live in now. I want '99 XJ world again.

 

[ericw.]

First of all, Chat GPT and other "AI" LLMs are nothing more than glorified text prediction.

That's not true at all and you even contradicted this in your final sentence.

Short answer, I'm 99% sure Chat GPT is full of shit and regurgitating some nonsense it cobbled together from clueless forum and Reddit posts.

If all it did was text predictions, it wouldn't parse hundreds/thousands of search results. It also attempts to qualify and predict accurate answers based on the knowledge base of the entire internet... Which is exactly what most people are doing when they're searching for answers.

It's a tool in the toolbox, not the solution. If you keep expecting it to be a solution, then you'll continue to have a bad experience.

 

[HYRYSC]

GROK is better

 

[Sweetums]

That's not true at all and you even contradicted this in your final sentence.




If all it did was text predictions, it wouldn't parse hundreds/thousands of search results. It also attempts to qualify and predict accurate answers based on the knowledge base of the entire internet... Which is exactly what most people are doing when they're searching for answers.

It's a tool in the toolbox, not the solution. If you keep expecting it to be a solution, then you'll continue to have a bad experience.

Yes it parses all that information to create a statistical model that is expressed in the answer it spits out. If it crawls nothing but nonsense, you get nothing but nonsense output to your prompt. It's only a little bit better than back in the pre-wikipedia days you'd ask Uncle Jimbo a questions and whatever crap he'd heard down at the bar is what you'd get out.

The problem with AI is the lack of guardrails, especially on versions that are public-facing. Internal AIs in specialized fields are more easily limited in their datasets to stay within the guidelines of the business or industry. Public facing AIs are trained on whatever they can find online or illegally pirate, that means often repeated bullshit is amplified and given authority by the AI answer.

This is already causing a feedback loop of bullshit, as AIs are being used to write articles and news reports - so the AI crawls Reddit bullshit, publishes it, and the next AI (or the same one) then crawls that article which is now a reliable, citable source for the bullshit, and the process repeats. It's the Human Centipede of bullshit turned back on itself and accelerated to the speed of cutting edge processing. This type of feedback loop has always existed, but it was at least slow enough and had enough thinking people in the loop that it could be slowed or interrupted. This is an automate bullshit engine running at near-C speeds.

This very question is a perfect representation of how critical thought is being lost. Anyone with even a half-decent sense of what's going on could tell you "grease the slip yolk" is pants-on-head stupid. Now we have to sit here and argue about whether Chat GPT's answer is bullshit or not (it is), and whether or not it can be trusted (it can't).

At this point I'm so done trying to contain the stupidity of ChatGPT and other AIs. OP, if you trust ChatGPT, then go grease your shaft and see what happens - or maybe have the AI turn the wrenches for you. It amazes me that people who want to do it themselves immediately try to skip the first step of understanding the system they are working on, how it functions, and diagnose the problem through elimination of variables. Nothing but lazy shortcuts, might as well just hire someone to fix it for you if you don't want to understand what you're doing.

If the original poster trusted ChatGPT to be right, they wouldn't be here double-checking what it said. So they don't trust the AI - and that means don't go to the AI to solve problems or think for you; or if you do, don't come and complain about the answer it gave.

 

[ericw.]

GROK is better

Not for this type of stuff. Grok is decent at fact checking the media because it operates without the mainstream bias.

Diagnostics isn't Grok's strong suit.

Yes it parses all that information to create a statistical model that is expressed in the answer it spits out. If it crawls nothing but nonsense, you get nothing but nonsense output to your prompt. It's only a little bit better than back in the pre-wikipedia days you'd ask Uncle Jimbo a questions and whatever crap he'd heard down at the bar is what you'd get out.

The problem with AI is the lack of guardrails, especially on versions that are public-facing. Internal AIs in specialized fields are more easily limited in their datasets to stay within the guidelines of the business or industry. Public facing AIs are trained on whatever they can find online or illegally pirate, that means often repeated bullshit is amplified and given authority by the AI answer.

This is already causing a feedback loop of bullshit, as AIs are being used to write articles and news reports - so the AI crawls Reddit bullshit, publishes it, and the next AI (or the same one) then crawls that article which is now a reliable, citable source for the bullshit, and the process repeats. It's the Human Centipede of bullshit turned back on itself and accelerated to the speed of cutting edge processing. This type of feedback loop has always existed, but it was at least slow enough and had enough thinking people in the loop that it could be slowed or interrupted. This is an automate bullshit engine running at near-C speeds.

This very question is a perfect representation of how critical thought is being lost. Anyone with even a half-decent sense of what's going on could tell you "grease the slip yolk" is pants-on-head stupid. Now we have to sit here and argue about whether Chat GPT's answer is bullshit or not (it is), and whether or not it can be trusted (it can't).

At this point I'm so done trying to contain the stupidity of ChatGPT and other AIs. OP, if you trust ChatGPT, then go grease your shaft and see what happens - or maybe have the AI turn the wrenches for you. It amazes me that people who want to do it themselves immediately try to skip the first step of understanding the system they are working on, how it functions, and diagnose the problem through elimination of variables. Nothing but lazy shortcuts, might as well just hire someone to fix it for you if you don't want to understand what you're doing.

I don't know of a single AI model that actually gives weight to reddit posts and I've worked directly with ChatGPT and Gemini in a business role on the software engineering side.

Reddit receives the lowest score for relevance, and typically prioritized answers are where a problem was solved and a solution is available.

Add guard rails to your prompt if you want guard rails. That's also why giving it all the symptoms is important, telling it to only provide verified answers and the source, tell it to express doubt where there is uncertainty, and be blunt and to the point. Also, tell it to give you a number of possibilities, like the top 5 most likely answers/causes of the problem.

I didn't start the discussion about ChatGPT but I know this is a sensitive topic for you and every time AI is mentioned, you go off the rails on a rant. lol

Someone used it for diagnostics before coming to the forum. Not a big deal.

 

[Sweetums]

I don't know of a single AI model that actually gives weight to reddit posts and I've worked directly with ChatGPT and Gemini in a business role on the software engineering side.

https://www.cbsnews.com/news/google-reddit-60-million-deal-ai-training/

$60 million deal to have Google's AI trained on Reddit posts somehow slipped under your radar?

And don't even start on Grok, that is a perfect example of a whole different problem with centralizing critical thinking, how many times has Musk been caught manipulating Grok algorithms when it started saying things that he personally didn't like?

And yeah, I do go off about AI - because the way it's being implemented is dangerous and stupid. It's the usual Techbro "go fast, break stuff, damn the consequences" approach - but this time it's affecting everything from global economies to the ability of people to have deductive reasoning or critical thinking skills, to the society-wide perception of what is or isn't real.

Here we are in a thread where the AI answer is clearly bullshit and people are still defending the accuracy of AI.

 

[ericw.]

https://www.cbsnews.com/news/google-reddit-60-million-deal-ai-training/

$60 million deal to have Google's AI trained on Reddit posts somehow slipped under your radar?

And don't even start on Grok, that is a perfect example of a whole different problem with centralizing critical thinking, how many times has Musk been caught manipulating Grok algorithms when it started saying things that he personally didn't like?

And yeah, I do go off about AI - because the way it's being implemented is dangerous and stupid. It's the usual Techbro "go fast, break stuff, damn the consequences" approach - but this time it's affecting everything from global economies to the ability of people to have deductive reasoning or critical thinking skills, to the society-wide perception of what is or isn't real.

Here we are in a thread where the AI answer is clearly bullshit and people are still defending the accuracy of AI.

This investment doesn't mean that Google is adding weight to reddit answers. It actually states in the article that Google is only using Reddit to expand it's understanding of human language. That's not even close to what you're accusing it of doing.

The "search" enhancements are strictly to Reddit's benefit who will be allowed to use Gemini to expand it's own forum search capabilities.

I didn't say the answer was accurate. The answer isn't what's being defended. There is nobody that can tell OP what his problem is exactly. Can you? You can tell him what it's not, but you cannot tell him what it is.

We can give OP a list of things to check, and maybe it will be one of those things. That's what ChatGPT did and that's how the AI answers should be understood. Just something to check.

 

[Sweetums]

The Chat GPT answer is like saying "my electricity is out" and it spits out "flush your toilet". It's nonsense.

 

[ericw.]

The Chat GPT answer is like saying "my electricity is out" and it spits out "flush your toilet". It's nonsense.

You can search for yourself and find cases where a dry slip yoke has caused clunking noises. There's even a bunch of DIY videos across many makes/models on how to fix this clunk with grease.

It's not quite the extreme abstract suggestion that you're inflating it to be.

 

[red/green hawk]

Clunking under acceleration is very unlikely to be related to the slip yolk, this part telescopes in and out to allow the axle to move up and down. When the axle droops, it gets farther from the transfer case and thus the drive shaft has to get longer. When it compresses the distance between the transfer case and diff is shortened. Think of it as a right-triangle. The chassis is straight, and you measure straight down to the diff. The drive shaft is the hypotenuse, if the leg of the triangle that is the distance from the axle to the chassis is lengthened, they hypotenuse (drive shaft) also has to lengthen.

Since the slip yolk only slides in and out, there's not really any lash in this part of the system. Even if there was, grease won't fix lash, which is due to bad adjustment or wear. Clunking could be a number of things from bad bushings in the control arms to worn U-joints in the drive shaft, or excessive wear or poor adjustment inside the differential. It could also be worn out engine mounts, as the engine's torque pushes it around in the engine bay, but that's unlikely given the young age of our trucks.

I've experienced worn U-joints in other vehicles, and the clunking is nothing like that. Besides if it was the U-joints would there be vibrations and repeated clunking. This is only occaisional that I'm experiencing.

First of all, Chat GPT and other "AI" LLMs are nothing more than glorified text prediction. They use complicated math to predict what series of words are most likely to be used related to the prompt. The rush to AI-all-the-things is happening at the expense of reality, be very cautious about trusting AI for factual information.

Short answer, I'm 99% sure Chat GPT is full of shit and regurgitating some nonsense it cobbled together from clueless forum and Reddit posts.

Okay I get that sentiment, but it diagnosed my problem with the Insane Audio head unit with 100% accuracy. I've been searching a solution for that on forums and reddit for a long time before ChatGPT with no success. It can't be purely luck that it was able to tell me exactly what was wrong.

 

[red/green hawk]

The Chat GPT answer is like saying "my electricity is out" and it spits out "flush your toilet". It's nonsense.

It asked me very specific questions, and I gave it very specific answers, and it helped rule out why it might not be U-joints and or Control arm bushings.

At first, I was worried it was telling me what I want to hear, but I asked it questions from all different angles and answer ends up being the same.

Sponsored