ChatGPT to Diagnose Tacoma Issues

Has anyone used ChatGPT to diagnose engine problems? I’ve been messing around with this and it seems to be pretty informative for the most part, at least with older Tacoma models. What do you guys think about this?

 

No. I'm not signing up for anything else.

I have played with Google Gemini 2.0 and Perplexity.ai
But not auto related.

 

You can use it for free… you only have so many questions per day. Check it out. No sign up necessary.

 

Haven't looked in over a year. I'll check it out.

Started with Copilot, then Perplexity, and just hoped over to google when Gemini 2.0 was released.

I just play around. Haven't used any of it productively.
At most, I find it "interesting".

 

Haven't tried
Always down to try something new

 

I'm a programmer of over 27 years. Code and technology was my life until I retired (or, was cancel cultured due to age, see it as you wish.)

AI code is written by humans, relies on human data, and is going to have the same fatal flaws as humans (see: Terminator.) I detest the AI movement. I'd never use AI to influence anything in my life, including diagnosing the Taco. It's just another way we learn to be more lazy and reliant on technology to do our "maths."

To sum it up in one sentence:

"Alexa, wipe my butt."

 

Nope no intrest in AI and older Tacoma trucks parts are difficult to find and they're expensive to keep running. Can AI remove rust and build new OEM parts that can't be found?

 

ChatGPT is just a glorified search engine. That said ChatGPT has convinced me to side with the Brotherhood of Steal in Fallout. It doesn't matter how benign and helpful a synth is in the game, it's dieing now. FOAD AI search engines!

I am also now OK with killing the kid in AI, the movie, and side with the bad guys in the movie The Creator.

 

The more we rely on computers to tell us what to think, the more our critical reasoning skills atrophy. I want to work through the problem myself.

 

It can't be less useful than starting a thread here before using google to search for the answer

 

If that's all you think it is you're using about .001% of it's capability.

 

I haven’t used it to ‘diagnose’ (yet) but that’s only cause my truck is still pretty new

I HAVE used it in the shop doing routine stuff, as a test. Example changing oil, rotating tires, etc on my wife’s Subaru and I’ve asked it for torque specs on the drain bolt, lug nuts, and for the size of the crush washer. Without having the full maintenance manuals and Subaru forums being a wacky world… yeah it helped.

I double checked the info once I had it by searching what it told me and bingo bango confirmed and it was correct.

I think it’s helpful to respond with ‘smart enough’ info that either confirms what you had a hunch on or allows you to take it and then confirm its responses elsewhere.

yeah, it’s good to an extent cause Google and other searches can yield so much damn noise and BS

 

Most of my AI interactions haven't been that great self check out at the grocery store usually requires a reset by a human. The inventory system I use does not understand chronological order at least not in the same order all of the humans use. I don't see any advantages in those circumstances. The only thing I encounter that works well are automated call systems.

 

Or the search function on this forum.

 

I haven't really used it other than what Gemini provides with my Google searches. It seems pretty good, but I have seen it come back with questionable information. I have worked in tech for 25+ years, starting as a developer. We used a "neural network" pattern in one project way back in the early 2000's, but that was using a small data set and was intended to just adjust itself over time and keep readjusting for changing conditions. The methods aren't really new, it's just been highly optimized and we have way more power to throw at it now.

AI works great when trained on verified facts, or when doing things that work by a structured set of rules. It's pretty good at writing now (proper grammar and spelling, so watch out - those phishing emails are getting better and better) and can write code, both of which follow a clear set of rules.

Where it has trouble is when it's trained with unverified information and has to sort through it. If there are conflicting answers, it either has to present both or choose. How does it choose? Unless the training includes weighting sources (like trust Wikipedia more than what someone on TW says), it's got to basically go with public opinion, which isn't always correct. Since these LLM's are ingesting stuff from everywhere, who knows what kind of crap and misinformation they are ingesting.

People here on TW and other sites are still promoting doing the chalk test to set tire pressures, so I ran a query just for giggles (what tire pressure should I run after switching to E-rated 285/75R16 tires). Thankfully ChatGPT did NOT tell me to do the chalk test and provided decent information on how to determine the proper pressure, but it did not provide a precise answer. It provided general guidance that seemed reasonable, and knew what the OEM recommended pressure was.

Either way, these AI tools don't really think. They sort through massive amounts of information that came from humans to provide the best answer it can. If humans stop thinking and providing new knowledge, AI will never get any smarter and could get worse. There are some really good use cases, but many of these use cases could be solved using a simpler solution that uses way less power.

 

, well said especially coming from an insider of this field.
Perhaps it’s a little too extreme, but I was taught there’s no free lunch anywhere.

 

Except Linux. Linux is still free (for now.)

 

"Machine says your shit's fucked up."

 

With my slow processing mind, it takes AI as just anew the ‘garbage in garbage out’ theory from the old ‘search’ in search engines.

 

Yep, have to give Linux credit; it’s a good samaritan thus far.