AI advice

[jordway]

Just had to share this story.  Some of you may remember my posts a few years ago about a starting problem.  I went way down a rabbit hole of thinking it was a immobilizer problem, but turned out it was just dirty electrical connectors (no big surprise there, but I was pretty green at the time).  I’m embarrassed to say that my laziness caused me to just hand it over to my mechanic to clean all the electrical connections.  I totally underestimated the time and effort needed to do that job, and ended up with a bill just north of $1k!

 

That solved the issue for about 3 years. But, after my first drive of this season, the issue returned.  I certainly didn’t want to pay $1k again to do something I can do myself, even if it is a long and tedious job. 

I use the deep thinking version of ChatGPT for my work.  Just to see what it would say and to perhaps focus in on the most likely connectors to be the problem, I fed it as much detail about the car and the starting problem that I could think of.  I asked for potential causes/solutions.  I did not tell it that I suspected a dirty electrical connecter.

It gave me a few possible causes, but ranked a dirty connector as the most likely problem.  It suggested that the connector for the crankshaft position sensor was the most likely problem.  But, it pointed out that every owner-built Caterham is a bit different, so it’s hard to really know (I am the second owner, not the builder).  It asked if I had photos of my engine that I could upload.  I have a set of 9 photos that cover every part of the engine.  I gave it those photos. 
 

It churned for about 15 seconds (a long time for AI).  It came back and said, “The photos helped a lot!  Your car has a MSD ignition system with a remote-mounted coil.  It’s much more likely to be the connector for the coil.  And here is one of your photos with the exact connector that you need to clean circled” (attached).

 

When I got home, I first tried starting the car a few times.  No start.  I then cleaned only that one connector it pointed out.  I’ll be damned if it didn’t start immediately on the first crank. 
 

That’s pretty cool.  I’m sure that I will need to clean other connectors, too. But, ChatGPT went further to identify all the components and their wirings paths, label my photos for all parts, connectors and ground points, and give me the order that I should check and clean them.  Awesome.   

 

 

[Kitcat]

Science fiction is becoming reality. 

[Vovchandr]

AI is very good at deciphering and summarizing large texts.

 

AI is also very good at parsing the internet for all solutions that people have used to resolve a problem in the past. Same thing most of us used to do by browsing 8 different forums with 15 different search queries and dozens of pages of people coming up with possible solutions. 

 

I use it frequently and its great at letting me skip the leg work of parsing for solutions from other people and just bringing them front and center. 

 

It poses a great threat for forums and Facebook sections where people look for technical answers. 

Edited April 1 by Vovchandr

[speedwagon]

No, AI is just a progression of what we had to do manually in the past. Trouble shooting is- data accumulation, and anilingus. In the 50s we kept equipment history records and used that to match the problem we were dealing with to the problems that had occurred in the past. We used to add, divide and multiply to solve complex math questions. today we can describe our problem and get answers as to what was done in the past to solve the problem. SAME process, just automated ( done by the cooperation of many machines that operate far faster than our "human" brains can.  So we are still at the old drawback-(garbage in=garbage out) you just have to be smart enough to ask the right question.  AI is definitely artificial but not as yet as intelligent as most of us are.  faster than lightning when the path is known, but unable as yet to blaze new trails.

[speedwagon]

as Vovchandr said. so well.

[jordway]

Yes, agreed, @speedwagon.  AI is TOTALLY a garbage in-garbage out thing.  The main issue is indeed understanding how to ask your question correctly.  It takes some practice.  I find that many academics are very resistant to using it because it's too easy and doesn't require the important grunt work of the past.  But, you know, I no longer go to the basement of the medical school library and dig out hard copy journals and photocopy the papers I need to read.  Why would I do that?  Same thing here.

[panamericano]

Now the master mechanics can retire.  The R&R technicians can get AI to tell them what part to remove & replace  Cleaning/repairing may still be a bridge too far for some.

RIP Phil Remington, et al.

[Vovchandr]

Just like one needed nuance before AI in parsing suggestions from strangers on the Internet to your problem one also has to parse what AI suggests but it's much easier to instantly work and adjust AI list then it is to deal with one way communication of 7 year old forum thread. 

 

It certainly is the best person to talk about Pectel information as resources for that are pretty much non existent. 

[speedwagon]

If we the purchaser would refuse to buy anything that did not come with a detailed  service manual containing-- assembly and disassembly guide, parts location, a complete parts list, electrical schematic, and calibration procedures,   all would be good. but we would not be able to buy anything at all. We live in a throwaway society today. Where the auto makers do not want you to be able to fix anything, Because the maker/seller partners make far more money servicing you than it cost to make the car. to-whit  The Front Lower control arms on my 2007 toyota highlander require (according to the toyota factory manual) THE REMOVAL OF THE ENGINE, and of course this is priced accordingly.    Now my friends just ask the question of AI as to just WHY is this so.   We are driving autos that were poorly designed and made, and being required to carry the burden of making the pp product usable.  Yes my car will go 100k miles, but I will be out $100k for "normal"maintenance.    we are being hosed

[wdb]

I recently used ChatGPT to suss out the answer to a fairly complex question relating to retirement finances. It was an iterative process -- I asked my initial question, then the chatbot came back with more questions, which I answered. On it went until a conclusion was reached. The answer was what I kind-of sort-of more-or-less thought might be the case before I started the process.

 

Now my question is: did my initial question, and/or the responses I gave to the bot's questions along the way, influence the outcome? Did I reach the answer I wanted, in other words.

[speedwagon]

                repeat the same process but use different answers to any secondary questions that you are not absolutely sure about. if you get different answers= then ask which are right for you-or use your own "I" to chose. Remembering that AI uses available data, and does not necessarily cull out false info, or invent new solutions like true "I" might. AI is just a tool that for most of the time is faster and more reaching.  Beats spending 5 days at the law library.

[jordway]

Another helpful trick is to ask it to give you links to the source information it used for a particular conclusion, especially for anything that seems suspicious.  You are then back to doing the digging yourself, but at least you've vastly narrowed it down.  In my experience, the out-of-whack answers are usually due to asking it a vague question.  For example, I was once asking for it to calculate mass balances for a particular process.  There are two ways of doing that particular process, and I didn't specifically tell it which way I was talking it about.  So, it gave me the average of the two, which makes no sense in the real world.  Once I was more specific, it nailed it.

[slowdude]

I've used it pretty extensively at work. Built an app which references a bunch of policy information etc. 

Think, I used a pre made a sandbox, defined the sand to play with in the sandbox. 

 

I've used it to summarize 30 pages of notes from when I started a job to build out a handbook for the role. 

 

Copilot can build heatmaps, adjust powerpoints etc.

 

I also used copilot to update and review my map for the coil on the emerald. it was able to figure out what the base values should be, and then compensate for voltage drop.

[Scott_]

1 hour ago, slowdude said:

I also used copilot to update and review my map for the coil on the emerald. it was able to figure out what the base values should be, and then compensate for voltage drop.

That's pretty cool.    There was an HP Academy video posted sometime back on youtube where the presenter took ecu tuning advice from AI with some interesting results, but my take away was so long as the input information and parameters were solid the results could be very useable like in your use case.  

 

As I am learning the Haltech software I've asked for input from chatgtp on settings and menu choices and it's been helpful as a learning tool.  It walked me through things like how to set up generic outputs  with no issues.

 

Scott

Edited April 4 by Scott_

[slowdude]

Yep, you will definitely need to know some of the limitations of the engine, and to know the timing, but i've found AI great, even for checking for spots in the map that could be an issue. It calculated injector scaling etc.

AI isn't an end-all, its only as good as the prompts which it is given.

[IamScotticus]

I wonder where we go when the answers found using AI are taken as absolute and infallible truth, regardless.

 

Also, AI does not have tact (neither do I).  The truth can hurt a lot of feelings.  The truth can make people snap.

Useing AI as the weapon of proof that someone is inferior or a belief is deleterious to a peaceful society thus deserving of disparate treatment, the civil wars will ignite.