Been in the automotive business my whole life because I grew up with it. Every manufacturer can make garbage but more in the last 15 years. From the KIA/Hyundai 2.4 Theta, the Ford 6 liter and 6.7 (Edit 6.4) liter diesels, Gm 1.4, 3.6, 5.3 with DOD... It goes on and on and people will defend them with "well mine has 83K miles and it's been pretty reliable...". Every vehicle today should have NO trouble going 200K with scheduled maintenance. It's probably just an amazing coincidence that any engine with 'ECO' in the name is delicate garbage that will save you fuel expense in the beginning but screw you later when the Turbocharger(s) or head gasket(s) fail. And they will.
It seems like you have a strong opinion about the automotive industry and the quality of vehicles produced in recent years. While it’s true that every manufacturer can produce some less reliable models, it’s important to note that automotive technology has also advanced significantly over the years. Many vehicles today are designed to last longer and can reach high mileage with proper maintenance. That being said, it’s understandable that you have concerns about certain engines or components, such as those labeled with “ECO.” It’s always wise to research and consider the specific models and their track records before making a purchasing decision. Regular maintenance and following the manufacturer’s guidelines can also contribute to the longevity of a vehicle. Ultimately, opinions about specific manufacturers and models may vary, and personal experiences can influence perceptions. It’s essential to weigh various factors and make an informed decision when choosing a vehicle.
@@JamesCouch777 they know what they are doing. It's planned obsolescence. If they build dependable long term the public won't be buying new vehicles. It's a science to have powertrain failure right past the warranty period.
One day ChatGPT may be able to diagnose the issue, I doubt it will ever manage to do so with just the right amount of sarcasm though... I think you're safe for the moment Wes. Great video as always.
The fact that it did steer you in the right direction was pretty amazing. It's not wrong, it just wasn't exact. It also told you it wasn't going to be exact.
Yeah, it’s great at pointing you in the right direction, and giving you an idea of where to start diagnosing an issue. Obviously with just a web based text interface it won’t be able to be specific, but once you do get GPT4 like models with the ability to interface with the cars systems it will probably end up being a huge time saver for a lot of shops.
@@dos_gringos9853it won’t save time, (my personal guess) is that it just make more cars come through the shop for cheaper and eliminates the need for highly experienced workers. Not really a good thing for the mechanics.
He didn't follow up either. A good follow up prompt would be "thanks for the generic information but in all of the information you have stored what is the most common cause of these problems and how did mechanics fix it the most efficiently".
I’ve used it for diagnosing circuit boards and finding smc parts that didn’t have identifiable numbers on them. It’s a really powerful tool to get you going in the right direction when it comes to sorting through masses of data for components that meet certain criteria.
Sorry, but this guy is not a real mechanic. How do I know this? A real mechanic would have to look down at the key board to find the letters while typing. This guy didn't have to do that!! :)
The thing to remember is chat gpt is a general chat program trained on general information online. You could train an ai on specific data sets that are updated to include bulitons recalls and previously recorded correct repairs. This would allow it to keep updated with new and old vehicles as well as learning as it goes what repairs were successful m
Yep you can already use GPT4 (the engine behind ChatGPT-4), with custom data sets, so would be trivial to do, if you had access to the data/books/etc to feed it with, instead of one mechanics knowledge, you would have one that combines millions of mechanics knowledge, esp. for edge cases that every mechanic is unlikely to run into for a specific model/brand.
@@michaelnjensen most large company's won't share data with something like chat gpt as there faya handling policy won't allow it. Would need to be locally hosted and bespoke to the contract
I'm pretty sure even current chat GPT may have some amount of knowlegde of these specific issues. If there are news articles, forums or official info from the company regarding very common issues, you can ask GPT and it'd likely know about it.
@@TheRedRaven_ true but also not true. It can do basic code for you but it can't actually make any of the backend of a program work. Most of what it can do you could also find googling for yourself. My point is that if you made an AI program that was trained and centered around program for example it's ability's for coding would be far and beyond what chat gpt can achieve.
Taking the longer view, the fact that every mechanic knows that those engines typically have the same problems seems like exactly the kind of knowledge base that an AI diagnostic system would be perfect for. Perhaps the issue is that the ones currently available haven't been trained that way (yet).
I'm a software developer and I use AI to generate "scaffolding", or code that is self contained and limited enough I suspect AI will be used in many sectors the same way - near-menial tasks that require some intelligence, but just don't have the context of the wider situation. I reckon the main barrier to AI will be providing context/state, without having to type everything in. In this example, it can only give general information, and only knows what's it's been fed by humans
I'd think the big challenge is to give it enough intelligence that it can filter it's training set. For example, this is a Chevy 1.4. Include cars with the same engine and exclude the ones that are different makes.
This is exactly right. A lot of this info is not publicly available on the internet, and most of what this bot was trained on is just that, publicly available information on the internet. With stuff like automotive repair, as Wes said, even if it is available in a forum or something, there's a ton of complete trash information out there that you have to sort through, so by no means is it comprehensive. Get a mechanic who's worked on a bunch of these to talk to it for a few days, then we might have something.
@@randr10 I would take it one step further and have the AI access dealership repair records. IMHO, the AI can quickly see which models are coming back to the dealership and for what types of issues. Share this information for public consumption i.e. the people who turn wrenches in the field.
Here's a link to the ChatGPT query: chat.openai.com/share/e4f50a10-164e-46bb-82d9-0841f6731131 Thanks for watching and special thanks to the folks who support us on Patreon at patreon.com/watchweswork
I am using it right now, it is very good a lot of different things. It can give you detaied step by step directions on how do a service, or bake a cake.
To me the ChatGPT suggestions would have been super helpful! I know very little of engines, but I can follow instructions. Cross referencing the instructions and googling the found symptoms would have guided me on the right path.
@@louf7178 I solved a vaccum problem in my house water system using GPT4, but thats not a car. Was caused by difference of pressure from mains caused by an improper pipe fitting. No more pipe humming and water hammers.
The feet of the Orang Asli people are well adapted to running up and down trees. Our ancestors had rock-hard feet.The feet of people who live in modern environments are much different and much softer because we wear shoes. Agreed that we are getting "soft" and reliant on technology and we definitely need to proceed with caution - however the future humans, augmented by technology, will be capable of superhuman feats. I think it's quite exciting!
why do you cling to the natural world when its as flawed as the artificial fear change for no reason anyways, ai brings no greater intelligence because its just based off humans anyways. itll mimic the mistakes and worldly desires of humans that limit us. nothing changed except now we can mimic a dumb human brain thats just faster
That was a very clever way of integrating ChatGPT and quickly showing its limitations in this application. The uprising of the machines is not yet around the corner!
What people don't seem to understand is that chatGPT is simply a tool. It's similiar to referencing a manual / other help guide except being able to provide context to narrow down the search to the right answer. It's the same with programming, people assume it's just able to write code from scratch, but you need to know the right questions to ask and also understand that it's dataset is quite limited and a lot of the stuff it provides could already be outdated compared to the documentation provided. But again, this is more of a tool that can help you get to the right answer. Similar to how google is used as a tool to find the answer you're looking for, it gives you suggestions based on the search you type in, which it then has a bunch of websites indexed and performs a keyword search based on what you typed and what it has indexed and ratings based on a bunch of web crawler metrics to rate the websites and then display them in that order. The supposed "uprising of the machines" is inevitable so long as tech keeps advancing, which it will.
Wes, your opinion and summation of AI-ChatGPT is the most succinct I have seen or heard. I hope you make lots of money doing what you do so we can enjoy future commentary, no matter the subject. Your humor keeps getting better and better.
Your analysis of ChatGPT type A.I. is spot ON. It simply takes pre-searched data and reforms it as more human dialogue. An internet search by a human could come up with the same data, but of course, not as quick.
It can also tabulate the occurrence of data. The problem with Auto diagnostics is that much of that data is not public data except for what you find on car enthusiast sites and even then there isn't a large collection of cars, codes, problems with the solutions for chatgpt to tabulate and say "with those codes and car there is a 98% chance it's the O2 sensors." Other data yes, it can do that, but the automotive repair world keeps a lot of that data private as to protect the market.
Wes is genuinely a real world "MacGyver". Is there ANYTHING the guy can't fix? If so the list is very short :) Two thumbs way up high to Wes and guys like him!
@@michaelstrafello7346very true. It’s like the contractors house is falling apart because he’s always working on someone else’s house. I agree, this guys skill level is insanely genius. I’m yet to watch a video where he can’t fix something. Great job Wes as usual.
I have been using chatgpt regularly to ask questions and also help me with better phrasing some mails and I am very impressed by the results. It is a game changer for me.
@@wyattyeary8050 Not really b/c I still come up with the original text and ask the engine to rephrase it. It cannot create the content for me even if I pass lots of hints.
ChatGPT, like the internet commenters it was trained on, always seems 100% convinced that it's right, and even when it's making stuff up it still "speaks" as though it's sure it's telling you the truth. Great video Wes, glad that AI isn't taking our jobs any time soon 😂
Seems AI falls for the Dunning Kruger Effect quite easily. The problem is, as soon as AI is trained on something on a topic, AI may not know how much or how little it really knows on that topic, while we hopefully realize that we just scratched the surface and there is more to it. Sometimes people knowing only very little use what little they know with more confidence than a real expert would show, that is what the Dunning Kruger effect describes. If we don´t fall for it, as we know that we only have limited knowlede on a subject, that is what makes us question our decisions to a degree and makes us learn more on it. As AI has data, but no information on how accurate and complete that data is, will answer in full confidence. You may just get a preprogrammed warning regarding the limited database. And this is something AI might possibly never learn, as it can only work on data it has. if it gets more data, it gets wider "knowledge" but AI will never question its dataaset. Otherwise AI would have to question every answer it gives, not knowing what more might be there. In a best case scenario AI can try a prediction on the reliability of the answer, based on how much different sources it knows and what the variation among those sources is or if there are contradicting sources after all. Also For a lot of things I do not think AI will take our jobs very soon. Maybe for the one or the other Agent on a support chat, possibly. but when it comes to industrial production, as long as AI cannot build a factory and build robots to make products, or to build houses or do plumibing, electric, HVAC work etc, I don´t think AI will replace a grat number of jobs. I would more be worried about jobs in fields like Accounting, controllling etc, but then, who will rely soelly on AI without controlling AI. I see AI and its capabilities more like a tool we may use in the future and that for sure at some point will change how we work. But it will not replace our work.
It's been programmed and trained by humans, it's not really a surprise it behaves in a similar fashion. Those with neither intelligence nor education shout very loudly, whilst those with both stay silent. Pretty much the opposite of what's needed to make existence bearable.
I'd say it started to diagnose it. I think for a junior technician, it could be useful act as an intermediate step between asking another tech, and trying to _just_ google the answer. Also, ChatGPT 4 is quite a bit better than 3, IMO. I don't know if that would translate to being better at diagnosing something that isn't tangible to it, unlike code, but it would be interesting to see if it is.
It's good to see a mechanic. who is interested in other things other than twisting wrenches and you do a very good job of explaining what you're doing, how you doing it and reasons behind it. 5 stars
Hey Wes, I imagine a dedicated AI automotive diagnostic tool will come along pretty soon. Between live data capture, current fault codes, service data and TSBs, AI auto diagnostic tools could be a real boon. People will still be needed to find bad U-joints and broken exhaust hangers, but I bet half a dozen strategically located chassis vibration recorders gathering vehicle history could indicate a lot of those sort of faults too.
I agree - a very interesting line of thought. The data collected should be owned by the car owner too - so that s/he can exploit it and be rewarded for creating it and trading it back to the manufacturer or to other other vehicle owners directly.
@@IanGibson-c2w Why can't you be part of the solution rather than part of the problem? The internet was designed for free exchange of information, not as a profit making scheme. There are plenty of ways to make money from the internet without co-opting the freedom of information.
Actually not. The "dedicated AI" era is a past thing and it was declared dead in the day when Google unveiled the project "Transformer" that literally unleashed ChatGPT and other big generic AI. In short we have turned the page and we are living now the "generic AI" era and this is what allowed the quantum leap of AI in the latest 5 years because is exactly the fact of it not being "dedicated" that it allows it to be so smart. The counterpart is this is what turns the AI so dangerous for human civilization: very quickly it will become so exponencially smarter than us and will take the control. In fact it is already happening right now while you read this, in the minimal things that you consider harmless like your smartphone, your TV, your computer, your smartwatch, surveilance cameras, etc. In fact we are already trapped in this technology mousetrap and we stepped into it by our own will. In 5 years we won't recognize the world as we know anymore. The good part is that we don't have to concern about this because there is nothing that can be done to revert it: we already passed the return point a few years ago.
i disagree , ai cant do 90 percent of the labor , it isnt going to build an engine or transmission for you , its not going to diagnose any car older than 2012 which is what ALOT of people are still driving . How is Ai going to find an oil leak ? how is it going to know if you blew a head gasket??? how is it going to do a valve adjustment on a honda with feeler gauges that you literally need a humans sense of touch and depth perception/ tension to complete ??? there are alot of tasks that just cant be delegated out without completely changing how engines and literally the entire industry is built . and even then youll still have mechanics building all the custom stuff for all the enthusiasts and track people
I agree with the search engines have become almost useless because I get so much garbage every time I’m looking for something that I give up most times. Good video 👍👍👍👍👍
Wes, rest assured, once ChatGPT is being used at a mass scale, you'll start seeing more advertisements popping up when you're using the service. The internet isn't a dumpster fire, the entire planet is at the moment.
Chat GPT is already being used on an extremely wide scale. People are testing it in every field. For example I have tested it in process engineering knowledge and logic and sometimes it passes. Sometimes it lies through it's teeth. The worrying part is some people think it's infallible.
I know you Americans always think that the whole "planet" is your own country but rest assured the other side of the planet is not a dumpster fire like America right now.
@@DanielHatchman I don't think anybody think it's infallible, quite the opposite. I think most are concerned with bad actors using this for nefarious reasons!
First time watching your channel and i just gotta say.. love the way you presented this and you just got a new subscriber. I know ppl say this all the time, usually i think a lot of them are fake but i'm not. And i don't normally do this, seriously. My background is in IT but ive always had a fondness for working on cars in my teenage years and now in my midlife crisis it's something i'm getting back into. It never really crossed my mind to use ChatGPT (and friends) for something like this but after seeing the responses i'm quite impressed. I've got an old car with a few issue like this that i've been slowly fixing up and yeah ive mostly just been leaning on forums and stuff but they are hit n miss, but so time consuming to trawl through. The kicker though that earned the subscribe is that you seem to actually have a grounded view of what LLM's actually are and the pros/cons.. particularly that it's merely predictive and not factual, but as you've shown it can be useful to give suggestions or ideas that may actually lead someone with a little bit of knowledge to a solution... but you've still got to have some knowledge to be able to sort the fact from hallucinations in the responses etc. Essentially what i'm saying is that you've earnt my respect as i probably know more about LLM than you, but you probably know more about cars than me.. but the fact you've given ChatGPT a fair shake and have decent understanding it is what did it. Anyhoo, i've rambled enough.. i look fwd to binge watching some of your content over the coming days/weeks 😊
Had to do the exact same pipe swap on my daughters Trax, I swapped the pipe and O2 sensor all in one go and it went smoothly, new copper plated nuts and all. She says the car has a marked improvement in performance now that the leak and O2 sensor are working in harmony. Thanks for letting me look over your shoulder Wes. Cheers Sir!
I'm a mechanic myself and everyday I have to diag SOME kind of electrical problem on a vehicle and I've actually used some of the ideas, theories and processes that you do in your videos! You sir are a pioneer in the auto industry for self employed techs
I’m having an issue with signal voltage on an infinity I’ve ran a new circuit back to the ecu and replaced the ecu any chance you might be able to assist in where I should be looking next
@@Agilepickleunite not a trained mechanic, but are you sure the alternator works? It doesn't matter if an ECU is broken or not if it doesn't get the generated voltage needed to work with. Maybe check your fuse box as well. Make sure the power gets to the proper place first.
Thank you for all the work you put in to these videos and continuing to do so. I know this was a "1.1" hour job book time that you did in like fifteen minutes, but really you did it in like 5 hours, or more, considering all the filming, editing, thumbnail, uploading and etc... I'm just trying to say thank you, because we love this content, and I am a happy Patreon contributor because of it. You're inspiring, I'm doing the same, albeit part time, and I have a long ways to go and a lot to learn. Keep it up Wes.
Dude I bought my first car at 12 years old, that was 52 years ago. I trained in a technical college at 18. My first job out of school was at a Cadillac dealership in 1979. I was trained by a really old dude (like me), and also GM training facilities in Washington DC and Charlette, NC. All that being said, just to say, in all that time I have never seen a guy with such a grasp on how things worked. You should take that as a huge compliment, because I have seen some s**t over they years, and I am a perfectionist. We need more guys like you in the world! Yes I am way behind. No AI can not think like a human, and never replace us.
I think the key for chatGPT is understanding what it really is and was made to do. Like you said, it is a language model and the things it is good at make sense. It’s also really good at giving an answer that sounds good, even if it’s bogus, since it doesn’t actually understand the information. I’m doing some work on training an AI/whatever you want to call it model on some pretty specific technical stuff and it is tedious and requires a ton of work to even get the bare minimum training done.
Oh I work on automating code is way more involved for AI mostly because of computer resources you will need and the database needed to train the AI. I don’t think AI can replace humans, but that they can become great tools that can help humans learn a bit faster and become smarter, obvious,y if you know how to use it.
Well said to expand on what you said i heard a good explanation on AI while looking into the topic. "AI is good at saying what a solution should be not what it is"
I've watched you, Wes for years and I really felt the need to thank you for your effort to consistently make along the best content on UA-cam. You are a national treasure...
@@MindBodySoulOk Try formulating your questions in a more understandable way. There are, actually, companies that formulate "Prompts" that are more easily understood by AI systems. Formulating easily understood queries also helps when communicating with humans.
I honestly lost track at this point on how many times I have had to work on Chevy Sonics and the Trax for these the same exact problems, well over 20 at least especially those damn flex pipes... Illinois just doing what Illinois does with its road salt just blowing holes in our ships. At least that repair is a lot easier than what I'm currently doing, changing exhaust manifolds and gaskets on an 89 F150... I'm taking bets on how many bolts break lol. Definitely a cool video👍
I think this could definitely be helpful for independent shops. I'm a GM technician at a GMC dealership so i can just use the service manual to lookup those dtcs and follow a flow chart and lookup diagnostic tips easily. Definitely a cool concept, but i doubt it'll change much at least at the dealer level
Hi Wes, thanks for all of your videos. I may not understand most of what you work on, but your presentation and reasoning have helped give me confidence when I work on my vehicles.
Wes, I absolutely love the way you figure everything out. AI doesn't stand a chance with you around. Keep up the great work and wonderful videos. Thanks
I would argue this actually was really helpful. Someone who is wrenching at home with not too much knowledge may be able to diagnose some of the issues they have using GPT.
I tried ChatGPT at calculating the centripetal force acting on a person at the equator which is about half a pound and it got all the equations right but calculated the wrong answer five times in a row. One of the answers was many times the weight of the person so he'd fly into the air. I already knew the right answer from an online calculator so I just kept telling it to try again and it apologised and screwed it up again. It's very good at other stuff but it boggles the mind that the one thing a computer can't do is calculate.
That's because the way it works is just predicting the next word based on probability. If you use gpt4, it can do math because of a Wolfram alpha plugin.
The calculation cant work anyway because gravitational theory is no good. The earth cannot be spinning we would be in a centrifuge like environment where most light things are blown away from the surface of the earth.
What do I think? I think that it’s a bright spot in the week when I see a video from Wes. Always entertaining while being informative. No way in hell any AI will replace an original like Wes!!!
Good overview. I suspect over time, this technology will mature or get bought up and destroyed. The internet (dumpster fire) is a great example. Great, diverse content.
I tried to prompt it to give me the best steps to diagnose this issue based on the codes to minimise time and money spent on diagnosis and it basically gave me similar steps. I think because replacing the Cat is expensive and time consuming so it had some sense of that as a last resort. Which isnt totally irrational. 🤷♀️ This is the output I got: The combination of error codes you provided for your Chevy Trax can indicate a few potential issues. Here are the most likely causes and the steps you can take to diagnose and minimize the time and budget spent: 1. P0171 - System Too Lean: This code indicates that the engine is receiving too much air relative to the fuel, resulting in a lean air/fuel mixture. The most common causes for this code are a faulty Mass Air Flow (MAF) sensor, a vacuum leak, or a clogged fuel injector. - Start by inspecting the air intake system for any leaks or loose connections. Check the intake hoses, clamps, and the air filter box. - If there are no visible leaks, the next step is to clean or replace the MAF sensor. It's located between the air filter and the throttle body. - If the issue persists, you may need to have the fuel injectors professionally cleaned or inspected for clogs or malfunctions. 2. P1101 - Intake Air Flow System Performance: This code typically indicates a problem with the Mass Air Flow (MAF) sensor or the Throttle Position Sensor (TPS). - Since you already checked the MAF sensor while diagnosing P0171, verify that the sensor is properly connected and clean. If necessary, clean or replace the sensor. - Additionally, inspect the wiring and connectors related to the MAF sensor and the TPS for any signs of damage or corrosion. 3. P0420 - Catalyst System Efficiency Below Threshold (Bank 1): This code suggests a problem with the catalytic converter or the oxygen (O2) sensors. The catalyst's role is to reduce emissions by converting harmful gases into less harmful ones. - Before jumping to conclusions about a faulty catalytic converter, check the O2 sensors first. There are usually one or two sensors located before and after the catalytic converter. - Inspect the wiring and connections for these O2 sensors, ensuring they are in good condition. - If the sensors are functioning correctly, then a failing catalytic converter may be the issue. Consult a professional mechanic for further diagnosis and potential replacement. By following these steps, you can cover the most common causes for the combination of error codes you mentioned. Remember to proceed with caution and consult a qualified mechanic if you're unsure about any steps or if the issue persists.
You so right about needing a confidence level. I’ve asked it to do that and it kinda worked for that one thread. Also it just a matter of time before Chat gpt results are cluttered with ads and sponsored links.
I've been working in IT for over 35 years now and lots of new tech has come and gone in that time. Some of it was great, some of it was garbage. AI is the first tech in all of that time that actually scares me. People are trampling all over each other to implement it while AI experts who have spent their entire careers working on this are pleading for people to stop. This isn't going to end well. As for the Internet in general not all of it is a dumpster fire...there are still plenty of options out there in terms of software, browser add-ons, privacy-respecting sites, etc. that you can use that eliminates 95% of the garbage.
Wow. BS much? No one has been in AI their whole career - unless their career is so short as to disqualify their opinions. No-one is pleading for it to stop. Elon musk has made a few comments, but considering he is the dumbest smart guy out there, who cares? Anyone who thinks moving to a planet w/o an atmosphere (to speak of) is a viable plan has zero common sense. every accident will result in death death death. dumb as the clown going to 4,000 meters with a window rated at 1,000 meters. DUMB! this is the same guy who spent 44 billion on twitter which makes almost no money. DUMB! just so he can piss off the libs - who are the main potential customers of his electric cars. DUMB! Sam Alman? Where do I start? WHACK!
Thanks for questioning the hype! Since you had already found the issues it was interesting to see how AI approached the problem. Since we are stuck in the age of truthiness it's refreshing to get some factual feedback and a reality check! Great video!
Hey Wes keep up the good work I love watching your videos even the electronic billiard videos cuz it kind of helps me sometimes if I have a wiring problem I love your videos Man
Studied AI (well, neural networks really) in 1997/8 and from what I can see, nothing has changed about the tech in any major way, but now with the internet, it has access to untold amounts of data. Because it's machine-based, it can remember all that stuff. So it's basically a big rememberer
Just wanted to chime in on my VERY narrow experience with ChatGPT. I use it every single day. As you said, it's fantastic with programming. Its ability to convert natural language to code and even integrate your own variables is fantastic. I've also learned some new PowerShell techniques. I come from a long line of scripting languages (20+ years of large-scale Identity Management and MS AD experience), so getting rid of older bad habits I've picked up from batch programming, to VBScripting, to Python, and now Powershell. It's been great with most of the war history stuff I throw at it as well. I think your assessment is spot-on. Appreciate your multi-disciplined perspective. Like I mentioned, mine is super narrow. Cheers, brudda.
Someone asked chat gpt what color blue was, and got some strange things for results, like "blue is a primary color", or "blue is a color in the visible spectrum", but it also said it didnt understand the question.
I agree with you on the AI issue, not right now. It was a big todo in the machine manufacturing industry too (machinist) however it has come closer in that arena than any other with CNC mills, lathes and grinders with automated pick and place equipment. Thanks again and, keep on smiling.
I just worked on one of these for a friend and I had all those same codes and it wound up being the PCV valve stuck open but since it's molded into the valve cover it wound up costing quite a bit more that it should have
I feel, like a shade tree mechanic (or a young person just learning) might find these suggestions useful. I have no doubt that every mechanic knows lots of things, but I don't believe there's a public-ally accessible database of those things, and I don't know if all the TSBs are out there either. I'm going to say, that based on other things ChatGPT has answered, those answers were actually pretty good.
Totally agree with your assessment of ChatGPT. I was initially pretty freaked out by it but as I've learned more about its limitations I've become less and less impressed with it.
Just discovered your channel a few days ago and have been binge watching. Love it! Also: 1995 Unicorn dodge. Change the steering shaft! Everyone overlooks it because it’s hard to determine how worn it is. Trust me, I am 100% positive the wondering is due to a worn steering shaft! You asked for opinions from people who know… Brother, I know! Thanks for the videos!
I’m a repair guy and I use chat gpt all the time. It’s a powerful tool. You have to know how to roll the dice, and reroll. I ask the same thing four ways with different qualifiers. Super helpful for teasing out issues.
Hey, Wes from the south here! Great job in explaining and showing everything. I do these valve covers all the time on euro cars. I agree with you about the AI doing diagnostics but it will never replace an experienced tech. I bet if the AI had all service information and TSB's it would work better rather than random web answers.
I’m so glad that you referenced that the fact that chat gpt is merely a indexed tool that simply goes through all the data it’s been fed and literally finds the most probable answer. Another great video with great explanation 👍
that ABSOLUTELY is NOT AT ALL how LLMs or ChatGPT works. not even a little bit. it is not intelligent or sentient, but your description is completely wrong. During training, the model is shown a sequence of words and taught to predict the next word. The model assigns a weighting to each part of the input data based on its significance and changes the weightings based on the difference between its prediction and what the next word actually is. This process of self-supervised learning is repeated until the model reaches the acceptable level of accuracy before an evaluation is carried out using a test dataset that was not used during training.
@@jasonbender2459Exactly, in other words - it is not retrieving full sentences/paragraphs like a search engine, it's generating unique sentences depending on how you ask the question. For anyone who wants to see proof of this, try asking it a question that should modify the answer in a way that doesn't exist anywhere on the Internet, like 'Tell me the testing process for these codes, but from the perspective of a time traveller from the past who has never seen a car before'
When testing AIs they make sure the test data is NOT in the training data. That's called contamination. Yet the best AIs are out scoring human experts. Indexing? Not at all what's happening.
Excellent summary at the end, Wes. Keep in mind that very few people are as analytical as you are, so AI will (when used properly) assist the general population to think through problems and find solutions. But it is only in its infancy. It will almost certainly develop to a point in five or ten years to where we won't even remember its current limitations.
A couple years ago my 1994 F150 developed an issue that was an absolute pain to try and figure out. Took *months* of on-and-off diagnosis to figure it out, but I eventually did. After watching this video, I decided to try this out myself, to see if ChatGPT could have saved me any time or if it really was as hard to diagnose as it felt. I fed the following symptoms into ChatGPT: 1. When started from cold (i.e. after it's been sitting for more than 8 hours) or hot (i.e. if it's been sitting for less than 20 minutes) it starts up right away. However, if I try to start it at any point between those two times, it cranks for an extended period before it starts. If I push the accelerator all the way to the floor it'll usually start up fairly quickly. 2. While it's warming up and the computer is in closed-loop mode, it runs just fine. But as soon as it enters open-loop mode, it starts misfiring like crazy at idle. 3. The misfire ONLY happens at idle. When I'm giving it throttle to either accelerate or maintain speed, it runs perfectly smoothly. It only starts to misfire once you come to a stop and the idle drops. 4. The misfire is very evident if you stand by the exhaust. You will periodically hear a "chuff" when a misfire occurs. It happens several times a second but the spacing between them is not consistent. 5. The exhaust smells of gasoline, which to me indicates that it's running incredibly rich 6. When I run a KOER (Key On Engine Running) test, I get diagnostic codes 173 (Lambda sensor too rich) and 70 (cylinder #7 failed cylinder balance test). CGPT suggested, in order, that I check the fuel pump, filter, and injectors, the spark plugs and wires, the ignition coil, the distributor cap and rotor, and the O2 sensor. My dad and I had tested and/or replaced all of these things (except the fuel pump, because it happened on both tanks so it was very unlikely that both pumps were failing in the same way at the same time). I then opened up a new instance to clear out CGPT's memory and fed the same symptoms in, but omitted any mention of a KOER test and the DTCs. The *very first thing* it suggested I do was check to make sure that the fuel pressure regulator wasn't leaking fuel into the intake. That was what ultimately had fixed the truck in the first place, and it makes me feel better about how long it took to figure out. Goes to show that just because a car is computerized doesn't mean the computer will give you helpful information (particularly OBD-I. I've seen toasters smarter than early ECUs). I chased my tail for MONTHS because of those stupid trouble codes.
I agree with you wholly about AI. I prefer reality, not plastic, artificial crap. Now I'm going to saddle a horse and check out a good trout stream to use my bamboo fly rod in . How's that for reality ?? I like watching you figure out stuff and fixing it right .
I've been following technological trends since I read Ray Kurzweil's book Age of Spiritual machines around 2001 or so. I can say that we're not at the AGI stage yet, even with GPT 4, but the quality of this thing is quite the milestone that almost everyone said wouldn't happen within our lifetimes. As far as I can tell, we're a little behind schedule on Ray's predictions, but we're not too far off. I think he said we'd be able to buy a PC type device with human-level intelligence (or at least human brain computational capacity) by 2029. I think the trick they'll have to pull now is integrating things like Wolfram Alpha (which Steven Wolfram is actually working on) to fix all of the uncertainties that the LLM model entails. The next step will be miniaturizing and upping the efficiency of it so that it doesn't take up a whole warehouse to make one of these go. There's been a lot of work in the open source community on that front, to some success. Check out a fella named Brian Roemmele if you'd like to learn more. I think it's really interesting that the things it gets wrong are the same kinds of things that people are generally bad at. Its memory is murky, it's not good at arithmetic, it lies instead of admits it doesn't know or it's not sure, but damned if it doesn't talk just like a person. These flaws are all very human-like qualities, which was the thing that startled me the most. We think of computers as having this indelible perfect memory but turns out that if you stack millions of linear functions together as if they're neurons it acts an awful lot like a stack of millions of biological neurons and becomes terrible at math lol. The emergence of ChatGPT tells me that AGI will be here within our lifetimes. I figure another 5-10 years due to the exponential nature of these things. Now that they know what makes it go, they just need to optimize it, so it's only a matter of time. The incentives to get there first are astronomical too. This could end up being Pandora's Box or the greatest thing we've ever done. Only time will tell.
I'm a software engineer with a substantial mechanical background and ChatGPT has become an almost integral tool for me, despite the countless complaints I have about it. However I've gotta point out that (as you said yourself) it gave steps to follow (generic or not), it provided feedback on your responses (generic or not), and it did in fact lead you to numerous correct solutions. Also I think that the "it doesn't know all the known problems for this car" is simply just something nervous people would say. There's many mechanical forums online that ChatGPT has access to, but the one thing it doesn't have is bias... Hence why only people say "X cars are chunks of crap". You also had the privilege of seeing the car in person, and I'm keen how you think you'd compare to ChatGPT if you were to be diagnosing the car over the phone? For someone who's broke and not mechanically minded this could have potentially saved them hundreds of dollars at a dealership (if not more). So for what was meant to be an impartial review it felt a bit like a "tooting ones own horn" session to me 🤷♂️
I'm as old as dirt. I hate confusers but we are forced to use them just to make phone calls, access our money, listen to music, or watch tv. I've said it before Wes, I'll stick with carbs and points distributors-and Briggs and Stratton flatheads. After over 80 years, they still do the job and can be fixed and maintained by us mortals. To bad we can't get good parts anymore. Thanks for posting.
I have a 2013 Buick encore same engine, I replace the exhaust flex pipe with a new GM part, had no trouble with nuts coming off. My check valve failed in the intake, so I used a plastic plug and JB weld epoxy to close the hole permanently and replaced the valve cover. I also replaced the flex pipe from the intake to the turbo and cleared the codes p0420 and p0171 on the car and the codes never came back and the engine runs smooth. In my opinion you should have change the timing chain tensioner between the Cam phasers when you had the valve cover off, it is very common for the bolts that hold it down to come loose, that's why you hear a rattle in the engine. I can hear the rattle when you start the engine at 17:40. I replaced the tensioner and clean out the thread holes and used lock tight for the bolts. Great video though, I like how you explained the PCV system on this car, I seen a lot of other video's on this engine and you are the first to explain it.
Yeah, that's one thing I noticed about using google or any other search engine for that matter. It's just a bunch of crap popped up to you, so it definitely felt good to have something like chat gpt help in that regard.
The long term fuel trims usually take a couple of trips-- and/or a certain amount of run time for it to relearn and walk back down to the correct readings. When you see higher long term (such as 15%) and at the same time you see a negative short term (such as - 3% or - 7%, etc). That is the computer telling the long term setting that it's too high and needs to walk back down-- which it will do over time. Some cars learn very quickly and others have certain triggers that you have to hit first; like being fully up to temp and doing a complete trip, etc.
As a professional nerd, I've found that ChatGPT can sometimes be very helpful, saving lots of time searching for answers. However, sometimes it is totally wrong in its responses, and unless you know that, you can't call it out. I usually reply with "Are you sure?" with every response, and half the time it will correct itself.
Wes, a couple of things: 1. Prompting, or telling the AI what you want, is a whole skill in and of itself. If you give it generic "what do?" prompts, it will give you generic advice. I imagine you could have gotten a bit more useful advice had you told the AI your troubleshooting steps "Okay I went on a 20 minute test drive, here are the data points, I observed these, what are some likely causes?". You are right in that it doesn't have access to TSBs - but you can copy and paste the TSB into ChatGPT to give it some context as to what might be a good next step. It does have a good memory. 2. GPT-4 is an entirely different ballgame to GPT-3.5, which is what you used. It is hard to understate what a massive leap forward in intelligence and accuracy GPT-4 is. To put things in perspective. I don't know much Python, but I've told GPT-4 what I want, take that, run it, and iterate on that code going back and forth with GPT-4 until I have a functioning program. It is nothing short of miraculous. The only limitation is that it's 25 messages per 3 hours and it's slower than GPT-3.5, but it's great for complex problem solving.
I mean of course it won't steal your job, 90% of the work is physical, but the fact that it can help the layman diagnose the issue is really nice could you google most of this? probably, but you're gonna have to visit at least 5 different ad-ridden websites for all the info, here it's compiled into a neat list edit: I wrote that last sentence when I didn't fully finish the video, you actually addressed this exact issue, love it!
It seems like you had an interesting experience diagnosing and fixing a car issue with the help of Chat GPT. While AI can provide general information and suggestions, it's important to remember that it cannot physically inspect or diagnose a vehicle. However, based on the information you provided, Chat GPT suggested inspecting the intake system and mass airflow sensor, as well as checking for exhaust leaks before the catalytic converter. It also correctly mentioned that an exhaust leak downstream of the catalytic converter can affect oxygen sensor readings and trigger the P0420 code. You discovered an exhaust leak behind the catalytic converter and a vacuum leak in the PCV system, which you fixed by replacing the valve cover gasket and addressing the leak with a Quick Connect clip. You also replaced a faulty downstream O2 sensor. Regarding your thoughts on AI and its impact on jobs, it's an ongoing topic of discussion. While AI and automation can replace certain tasks and job roles, they can also create new opportunities and enhance productivity in various industries. It's important to adapt and upskill to remain relevant in a changing job market. AI, like Chat GPT, can be a useful tool in certain applications, but human expertise and judgment are still essential in many domains. Lastly, your observations about the state of the internet and the simplicity of receiving plain text answers from Chat GPT are valid. The internet can be overwhelming with ads, trolls, and information overload. AI models like Chat GPT can provide straightforward responses, but it's important to critically evaluate and verify the information obtained from any source, including AI models.
AI's prevalence and growing influence on our daily lives can indeed feel overwhelming, but it's important to remember that the progression and adaptation of technology has been a part of human society for centuries. While the rapid pace and complexity of AI can seem daunting, it also offers a variety of opportunities for innovation and improvements in efficiency. Just as past technological revolutions have reshaped industries and lifestyles, AI is poised to do the same. As with any technology, it's essential to be aware of and engage with the societal implications, which include job displacement, privacy concerns, and ethical dilemmas. Furthermore, a shift towards AI doesn't mean that everything becomes automated and human input becomes obsolete. There will always be areas where the human touch is crucial, especially in roles that involve creativity, critical thinking, and emotional intelligence. If the constant talk about AI is tiring, it may help to focus on learning about areas that you find most intriguing or relevant. For example, if you're a healthcare professional, you might want to learn about how AI is used in medical diagnosis. Or, if you're an artist, you might be interested in how AI is being used to create new forms of art. Moreover, public participation in AI discussions is vital. These conversations are not just for tech experts; they affect everyone, so everyone's voice is important. Remember that the goal is not just about 'surviving' the AI revolution, but shaping it to ensure it benefits all of society.
I use chatgpt for my programming projects and general questions. I know its answers arnt always correct but i like that i dont ever need to LOOK for the answer it just gives it me. So even though the answer it gives can sometimes be wrong overall it usually saves me a ton of time.
Very interesting evaluation. However, what was more enlightening for a former shade tree hack, is all the knowledge and equipment needed for just one manufacture's model and the Rube Goldberg plumbing required to make these things run with minimal pollution. Anyone who complains about the cost of repair would do well to watch this video.
Been in the automotive business my whole life because I grew up with it. Every manufacturer can make garbage but more in the last 15 years. From the KIA/Hyundai 2.4 Theta, the Ford 6 liter and 6.7 (Edit 6.4) liter diesels, Gm 1.4, 3.6, 5.3 with DOD... It goes on and on and people will defend them with "well mine has 83K miles and it's been pretty reliable...". Every vehicle today should have NO trouble going 200K with scheduled maintenance. It's probably just an amazing coincidence that any engine with 'ECO' in the name is delicate garbage that will save you fuel expense in the beginning but screw you later when the Turbocharger(s) or head gasket(s) fail. And they will.
It seems like you have a strong opinion about the automotive industry and the quality of vehicles produced in recent years. While it’s true that every manufacturer can produce some less reliable models, it’s important to note that automotive technology has also advanced significantly over the years. Many vehicles today are designed to last longer and can reach high mileage with proper maintenance.
That being said, it’s understandable that you have concerns about certain engines or components, such as those labeled with “ECO.” It’s always wise to research and consider the specific models and their track records before making a purchasing decision. Regular maintenance and following the manufacturer’s guidelines can also contribute to the longevity of a vehicle.
Ultimately, opinions about specific manufacturers and models may vary, and personal experiences can influence perceptions. It’s essential to weigh various factors and make an informed decision when choosing a vehicle.
@@WhiskeyGulf71 sounds like written with chatgpt
Same as appliances, sure it saves energy but after 3-5 years it's in the landfill. That's great for the environment.
You would think that by now they would have figured out how to make good vehicles but so many are just crap.
@@JamesCouch777 they know what they are doing. It's planned obsolescence. If they build dependable long term the public won't be buying new vehicles. It's a science to have powertrain failure right past the warranty period.
One day ChatGPT may be able to diagnose the issue, I doubt it will ever manage to do so with just the right amount of sarcasm though... I think you're safe for the moment Wes. Great video as always.
Agreed. I don't think the sarcasm can be replicated and that's what I'm here for.
Here here
@@buildingsalvage you mean "hear hear" right?😉
@@JoeInBendigo What ???
@@jimharper6073 why did you reply to me?
Does it relate to my comment to @M5?
The fact that it did steer you in the right direction was pretty amazing. It's not wrong, it just wasn't exact. It also told you it wasn't going to be exact.
Yeah, it’s great at pointing you in the right direction, and giving you an idea of where to start diagnosing an issue. Obviously with just a web based text interface it won’t be able to be specific, but once you do get GPT4 like models with the ability to interface with the cars systems it will probably end up being a huge time saver for a lot of shops.
@@dos_gringos9853Really though, I can point myself in the right direction with a Google search too, so...
@@dos_gringos9853it won’t save time, (my personal guess) is that it just make more cars come through the shop for cheaper and eliminates the need for highly experienced workers. Not really a good thing for the mechanics.
He didn't follow up either. A good follow up prompt would be "thanks for the generic information but in all of the information you have stored what is the most common cause of these problems and how did mechanics fix it the most efficiently".
I’ve used it for diagnosing circuit boards and finding smc parts that didn’t have identifiable numbers on them. It’s a really powerful tool to get you going in the right direction when it comes to sorting through masses of data for components that meet certain criteria.
ChatGPT aside, that oil leak was a great help in keeping those exhausts bolts from getting too rusty 😂
Haha total truth have a 335 in newengland and never had a seized bolt or nut
Sorry, but this guy is not a real mechanic. How do I know this? A real mechanic would have to look down at the key board to find the letters while typing. This guy didn't have to do that!! :)
@@trvman1☝️🤓 "sorry" it's called multitasking believe it or not many mechanics do that🤯
@@trvman1dude most mechanics do that if there making sure there typing the right thing or just don't know the keys
@@dynastycrew824i dont even have to ask if you are stupid..
The thing to remember is chat gpt is a general chat program trained on general information online. You could train an ai on specific data sets that are updated to include bulitons recalls and previously recorded correct repairs. This would allow it to keep updated with new and old vehicles as well as learning as it goes what repairs were successful m
Yep you can already use GPT4 (the engine behind ChatGPT-4), with custom data sets, so would be trivial to do, if you had access to the data/books/etc to feed it with, instead of one mechanics knowledge, you would have one that combines millions of mechanics knowledge, esp. for edge cases that every mechanic is unlikely to run into for a specific model/brand.
@@michaelnjensen most large company's won't share data with something like chat gpt as there faya handling policy won't allow it. Would need to be locally hosted and bespoke to the contract
I'm pretty sure even current chat GPT may have some amount of knowlegde of these specific issues. If there are news articles, forums or official info from the company regarding very common issues, you can ask GPT and it'd likely know about it.
Chatgpt can code software so I'm pretty sure it's beyond your basic google search.
@@TheRedRaven_ true but also not true. It can do basic code for you but it can't actually make any of the backend of a program work. Most of what it can do you could also find googling for yourself. My point is that if you made an AI program that was trained and centered around program for example it's ability's for coding would be far and beyond what chat gpt can achieve.
Taking the longer view, the fact that every mechanic knows that those engines typically have the same problems seems like exactly the kind of knowledge base that an AI diagnostic system would be perfect for. Perhaps the issue is that the ones currently available haven't been trained that way (yet).
I'm a software developer and I use AI to generate "scaffolding", or code that is self contained and limited enough
I suspect AI will be used in many sectors the same way - near-menial tasks that require some intelligence, but just don't have the context of the wider situation.
I reckon the main barrier to AI will be providing context/state, without having to type everything in. In this example, it can only give general information, and only knows what's it's been fed by humans
I'd think the big challenge is to give it enough intelligence that it can filter it's training set. For example, this is a Chevy 1.4. Include cars with the same engine and exclude the ones that are different makes.
@@hayleyxyz If you could plug it into the obd maybe.
This is exactly right. A lot of this info is not publicly available on the internet, and most of what this bot was trained on is just that, publicly available information on the internet. With stuff like automotive repair, as Wes said, even if it is available in a forum or something, there's a ton of complete trash information out there that you have to sort through, so by no means is it comprehensive. Get a mechanic who's worked on a bunch of these to talk to it for a few days, then we might have something.
@@randr10 I would take it one step further and have the AI access dealership repair records. IMHO, the AI can quickly see which models are coming back to the dealership and for what types of issues. Share this information for public consumption i.e. the people who turn wrenches in the field.
You are a smart man, my friend. You are keeping up with technology superbly. Used to be my passion. Not so much any more.
Here's a link to the ChatGPT query: chat.openai.com/share/e4f50a10-164e-46bb-82d9-0841f6731131
Thanks for watching and special thanks to the folks who support us on Patreon at patreon.com/watchweswork
I am using it right now, it is very good a lot of different things. It can give you detaied step by step directions on how do a service, or bake a cake.
To me the ChatGPT suggestions would have been super helpful! I know very little of engines, but I can follow instructions. Cross referencing the instructions and googling the found symptoms would have guided me on the right path.
But you probably never would have solved it. The bottom line is that people need to know what they're doing, IMO. It still might be helpful, though.
@@louf7178 I solved a vaccum problem in my house water system using GPT4, but thats not a car.
Was caused by difference of pressure from mains caused by an improper pipe fitting.
No more pipe humming and water hammers.
My fear of AI isn't that the "artificial" is so good, it's that the "natural" is getting so bad.
The feet of the Orang Asli people are well adapted to running up and down trees. Our ancestors had rock-hard feet.The feet of people who live in modern environments are much different and much softer because we wear shoes. Agreed that we are getting "soft" and reliant on technology and we definitely need to proceed with caution - however the future humans, augmented by technology, will be capable of superhuman feats. I think it's quite exciting!
Yep, the natural zombies on the streets of America.
The whole business model of the internet is convenience in exchange of privacy
@@dadikkedudethe human existence model has been based on convenience since writing and farming has developed lol
why do you cling to the natural world when its as flawed as the artificial
fear change for no reason
anyways, ai brings no greater intelligence because its just based off humans anyways. itll mimic the mistakes and worldly desires of humans that limit us. nothing changed except now we can mimic a dumb human brain thats just faster
That was a very clever way of integrating ChatGPT and quickly showing its limitations in this application. The uprising of the machines is not yet around the corner!
What people don't seem to understand is that chatGPT is simply a tool. It's similiar to referencing a manual / other help guide except being able to provide context to narrow down the search to the right answer. It's the same with programming, people assume it's just able to write code from scratch, but you need to know the right questions to ask and also understand that it's dataset is quite limited and a lot of the stuff it provides could already be outdated compared to the documentation provided. But again, this is more of a tool that can help you get to the right answer. Similar to how google is used as a tool to find the answer you're looking for, it gives you suggestions based on the search you type in, which it then has a bunch of websites indexed and performs a keyword search based on what you typed and what it has indexed and ratings based on a bunch of web crawler metrics to rate the websites and then display them in that order.
The supposed "uprising of the machines" is inevitable so long as tech keeps advancing, which it will.
GPT4 is not close to state of the art any more.
Wes, your opinion and summation of AI-ChatGPT is the most succinct I have seen or heard. I hope you make lots of money doing what you do so we can enjoy future commentary, no matter the subject. Your humor keeps getting better and better.
Are you an expert on ai? Or did he just say what you like to hear?
Wow! Some of your videos got almost 100 views. It's entertainment, DUDE.
Wes, I'm 72 years old and understand carburation and older American engines. I'm amazed at how much knowledge you have and your attitude.😮
Your analysis of ChatGPT type A.I. is spot ON. It simply takes pre-searched data and reforms it as more human dialogue. An internet search by a human could come up with the same data, but of course, not as quick.
It can also tabulate the occurrence of data. The problem with Auto diagnostics is that much of that data is not public data except for what you find on car enthusiast sites and even then there isn't a large collection of cars, codes, problems with the solutions for chatgpt to tabulate and say "with those codes and car there is a 98% chance it's the O2 sensors."
Other data yes, it can do that, but the automotive repair world keeps a lot of that data private as to protect the market.
That's not at all "all it does" it's not even what it does.
Wes is genuinely a real world "MacGyver". Is there ANYTHING the guy can't fix? If so the list is very short :) Two thumbs way up high to Wes and guys like him!
Probally his own vehicles because he never has time to because he's always fixing customer's cars😅in
@@michaelstrafello7346very true. It’s like the contractors house is falling apart because he’s always working on someone else’s house. I agree, this guys skill level is insanely genius. I’m yet to watch a video where he can’t fix something. Great job Wes as usual.
@@jasongreen5439you know what the say; the man,the legend 😊
Can we say he is near Chuck Norris category? No, and I reprimanded myself for even thinking it. I’m such and idiot!!
He's versatile and his approach and results are elegant.
I have been using chatgpt regularly to ask questions and also help me with better phrasing some mails and I am very impressed by the results. It is a game changer for me.
"The Devil's greatest trick was convincing the world that he didn't exist."
Takes the reward out of your work no?
@@wyattyeary8050bruh who cares about the reward, the reward is finishing faster
@@wyattyeary8050 Not really b/c I still come up with the original text and ask the engine to rephrase it. It cannot create the content for me even if I pass lots of hints.
@@wyattyeary8050 with that mindset every technological advancement we've had so far is pointless bc takes the reward out of everything
ChatGPT, like the internet commenters it was trained on, always seems 100% convinced that it's right, and even when it's making stuff up it still "speaks" as though it's sure it's telling you the truth. Great video Wes, glad that AI isn't taking our jobs any time soon 😂
Mechanics are often convinced they are correct until proven otherwise. But you have to keep asking yourself what you learned from being wrong.
Haha, you just summed up the computer version of a narcissist!!
lol yeah it's very relatable. People do that all the time!
Seems AI falls for the Dunning Kruger Effect quite easily. The problem is, as soon as AI is trained on something on a topic, AI may not know how much or how little it really knows on that topic, while we hopefully realize that we just scratched the surface and there is more to it. Sometimes people knowing only very little use what little they know with more confidence than a real expert would show, that is what the Dunning Kruger effect describes. If we don´t fall for it, as we know that we only have limited knowlede on a subject, that is what makes us question our decisions to a degree and makes us learn more on it.
As AI has data, but no information on how accurate and complete that data is, will answer in full confidence. You may just get a preprogrammed warning regarding the limited database. And this is something AI might possibly never learn, as it can only work on data it has. if it gets more data, it gets wider "knowledge" but AI will never question its dataaset. Otherwise AI would have to question every answer it gives, not knowing what more might be there. In a best case scenario AI can try a prediction on the reliability of the answer, based on how much different sources it knows and what the variation among those sources is or if there are contradicting sources after all.
Also For a lot of things I do not think AI will take our jobs very soon. Maybe for the one or the other Agent on a support chat, possibly. but when it comes to industrial production, as long as AI cannot build a factory and build robots to make products, or to build houses or do plumibing, electric, HVAC work etc, I don´t think AI will replace a grat number of jobs.
I would more be worried about jobs in fields like Accounting, controllling etc, but then, who will rely soelly on AI without controlling AI.
I see AI and its capabilities more like a tool we may use in the future and that for sure at some point will change how we work. But it will not replace our work.
It's been programmed and trained by humans, it's not really a surprise it behaves in a similar fashion.
Those with neither intelligence nor education shout very loudly, whilst those with both stay silent. Pretty much the opposite of what's needed to make existence bearable.
I'd say it started to diagnose it. I think for a junior technician, it could be useful act as an intermediate step between asking another tech, and trying to _just_ google the answer. Also, ChatGPT 4 is quite a bit better than 3, IMO. I don't know if that would translate to being better at diagnosing something that isn't tangible to it, unlike code, but it would be interesting to see if it is.
It's good to see a mechanic. who is interested in other things other than twisting wrenches and you do a very good job of explaining what you're doing, how you doing it and reasons behind it. 5 stars
Hey Wes, I imagine a dedicated AI automotive diagnostic tool will come along pretty soon. Between live data capture, current fault codes, service data and TSBs, AI auto diagnostic tools could be a real boon. People will still be needed to find bad U-joints and broken exhaust hangers, but I bet half a dozen strategically located chassis vibration recorders gathering vehicle history could indicate a lot of those sort of faults too.
I agree - a very interesting line of thought. The data collected should be owned by the car owner too - so that s/he can exploit it and be rewarded for creating it and trading it back to the manufacturer or to other other vehicle owners directly.
@@IanGibson-c2w more like a checkbox on the contract.
@@IanGibson-c2w Why can't you be part of the solution rather than part of the problem?
The internet was designed for free exchange of information, not as a profit making scheme.
There are plenty of ways to make money from the internet without co-opting the freedom of information.
Actually not. The "dedicated AI" era is a past thing and it was declared dead in the day when Google unveiled the project "Transformer" that literally unleashed ChatGPT and other big generic AI. In short we have turned the page and we are living now the "generic AI" era and this is what allowed the quantum leap of AI in the latest 5 years because is exactly the fact of it not being "dedicated" that it allows it to be so smart. The counterpart is this is what turns the AI so dangerous for human civilization: very quickly it will become so exponencially smarter than us and will take the control. In fact it is already happening right now while you read this, in the minimal things that you consider harmless like your smartphone, your TV, your computer, your smartwatch, surveilance cameras, etc. In fact we are already trapped in this technology mousetrap and we stepped into it by our own will. In 5 years we won't recognize the world as we know anymore. The good part is that we don't have to concern about this because there is nothing that can be done to revert it: we already passed the return point a few years ago.
i disagree , ai cant do 90 percent of the labor , it isnt going to build an engine or transmission for you , its not going to diagnose any car older than 2012 which is what ALOT of people are still driving . How is Ai going to find an oil leak ? how is it going to know if you blew a head gasket??? how is it going to do a valve adjustment on a honda with feeler gauges that you literally need a humans sense of touch and depth perception/ tension to complete ??? there are alot of tasks that just cant be delegated out without completely changing how engines and literally the entire industry is built . and even then youll still have mechanics building all the custom stuff for all the enthusiasts and track people
I agree with the search engines have become almost useless because I get so much garbage every time I’m looking for something that I give up most times. Good video 👍👍👍👍👍
Wes, rest assured, once ChatGPT is being used at a mass scale, you'll start seeing more advertisements popping up when you're using the service. The internet isn't a dumpster fire, the entire planet is at the moment.
Chat GPT is already being used on an extremely wide scale. People are testing it in every field. For example I have tested it in process engineering knowledge and logic and sometimes it passes. Sometimes it lies through it's teeth. The worrying part is some people think it's infallible.
I know you Americans always think that the whole "planet" is your own country but rest assured the other side of the planet is not a dumpster fire like America right now.
@@DanielHatchman I don't think anybody think it's infallible, quite the opposite. I think most are concerned with bad actors using this for nefarious reasons!
@@chrisstromberg6527 yes that's concerning too.
The ENTIRE PLANET is a massive wildfire
First time watching your channel and i just gotta say.. love the way you presented this and you just got a new subscriber.
I know ppl say this all the time, usually i think a lot of them are fake but i'm not. And i don't normally do this, seriously.
My background is in IT but ive always had a fondness for working on cars in my teenage years and now in my midlife crisis it's something i'm getting back into.
It never really crossed my mind to use ChatGPT (and friends) for something like this but after seeing the responses i'm quite impressed. I've got an old car with a few issue like this that i've been slowly fixing up and yeah ive mostly just been leaning on forums and stuff but they are hit n miss, but so time consuming to trawl through.
The kicker though that earned the subscribe is that you seem to actually have a grounded view of what LLM's actually are and the pros/cons.. particularly that it's merely predictive and not factual, but as you've shown it can be useful to give suggestions or ideas that may actually lead someone with a little bit of knowledge to a solution... but you've still got to have some knowledge to be able to sort the fact from hallucinations in the responses etc.
Essentially what i'm saying is that you've earnt my respect as i probably know more about LLM than you, but you probably know more about cars than me.. but the fact you've given ChatGPT a fair shake and have decent understanding it is what did it.
Anyhoo, i've rambled enough.. i look fwd to binge watching some of your content over the coming days/weeks 😊
Had to do the exact same pipe swap on my daughters Trax, I swapped the pipe and O2 sensor all in one go and it went smoothly, new copper plated nuts and all. She says the car has a marked improvement in performance now that the leak and O2 sensor are working in harmony. Thanks for letting me look over your shoulder Wes. Cheers Sir!
I'm a mechanic myself and everyday I have to diag SOME kind of electrical problem on a vehicle and I've actually used some of the ideas, theories and processes that you do in your videos! You sir are a pioneer in the auto industry for self employed techs
I’m having an issue with signal voltage on an infinity I’ve ran a new circuit back to the ecu and replaced the ecu any chance you might be able to assist in where I should be looking next
@@Agilepickleuniteask chatgpt
@@Agilepickleunite not a trained mechanic, but are you sure the alternator works? It doesn't matter if an ECU is broken or not if it doesn't get the generated voltage needed to work with. Maybe check your fuse box as well. Make sure the power gets to the proper place first.
You are not going to have to worry about being replaced young man. As always, good video sir, appreciate your time
Thank you for all the work you put in to these videos and continuing to do so. I know this was a "1.1" hour job book time that you did in like fifteen minutes, but really you did it in like 5 hours, or more, considering all the filming, editing, thumbnail, uploading and etc...
I'm just trying to say thank you, because we love this content, and I am a happy Patreon contributor because of it.
You're inspiring, I'm doing the same, albeit part time, and I have a long ways to go and a lot to learn.
Keep it up Wes.
I spent more like 25 hours making this video. That doesn't even count fixing the car. Thanks for your help!
@@WatchWesWork Och! Thank you.
@@WatchWesWorkI wonder if AI could work out the hourly rate you get paid for the vid
@@Brad.whatthe 🤣🤣🤣
@@GrayRaceCat Judging by Wes's comments regarding ChatGPT's lack of ability at math, I daresay it would get that wrong!
I enjoy your candor. Your no nonsense observations are refreshing.
Dude I bought my first car at 12 years old, that was 52 years ago. I trained in a technical college at 18. My first job out of school was at a Cadillac dealership in 1979. I was trained by a really old dude (like me), and also GM training facilities in Washington DC and Charlette, NC. All that being said, just to say, in all that time I have never seen a guy with such a grasp on how things worked. You should take that as a huge compliment, because I have seen some s**t over they years, and I am a perfectionist. We need more guys like you in the world! Yes I am way behind. No AI can not think like a human, and never replace us.
I think the key for chatGPT is understanding what it really is and was made to do. Like you said, it is a language model and the things it is good at make sense. It’s also really good at giving an answer that sounds good, even if it’s bogus, since it doesn’t actually understand the information. I’m doing some work on training an AI/whatever you want to call it model on some pretty specific technical stuff and it is tedious and requires a ton of work to even get the bare minimum training done.
Oh I work on automating code is way more involved for AI mostly because of computer resources you will need and the database needed to train the AI. I don’t think AI can replace humans, but that they can become great tools that can help humans learn a bit faster and become smarter, obvious,y if you know how to use it.
Well said to expand on what you said i heard a good explanation on AI while looking into the topic. "AI is good at saying what a solution should be not what it is"
I've watched you, Wes for years and I really felt the need to thank you for your effort to consistently make along the best content on UA-cam. You are a national treasure...
Can we make it global?
I have only gotten excuses from Chatgbt of why it can't answer my questions.
@@MindBodySoulOk Try formulating your questions in a more understandable way. There are, actually, companies that formulate "Prompts" that are more easily understood by AI systems.
Formulating easily understood queries also helps when communicating with humans.
I honestly lost track at this point on how many times I have had to work on Chevy Sonics and the Trax for these the same exact problems, well over 20 at least especially those damn flex pipes... Illinois just doing what Illinois does with its road salt just blowing holes in our ships. At least that repair is a lot easier than what I'm currently doing, changing exhaust manifolds and gaskets on an 89 F150... I'm taking bets on how many bolts break lol. Definitely a cool video👍
I think this could definitely be helpful for independent shops. I'm a GM technician at a GMC dealership so i can just use the service manual to lookup those dtcs and follow a flow chart and lookup diagnostic tips easily. Definitely a cool concept, but i doubt it'll change much at least at the dealer level
You fix a lot of p0016 codes?
@@BLYFACTOR not a bunch of that one in particular. I'm mostly doing 15-a and brake steering suspension
Independent shops have access to diagnostic information. At least, The decent ones that pay for that access
Hi Wes greetings from South Africa. Man it takes a special person to work on vehicles in the salt belt. You are a master mechanic!!
Hi Wes, thanks for all of your videos. I may not understand most of what you work on, but your presentation and reasoning have helped give me confidence when I work on my vehicles.
ChatGPt is also good at sounding super confident even if the information it provides is wrong
Basically like some HS teachers I knew.
Wes, I absolutely love the way you figure everything out. AI doesn't stand a chance with you around. Keep up the great work and wonderful videos. Thanks
What I've noticed using gpt is, for a detailed response, a detailed question must be supplied to the system. It's conversational. Cool video 😎
I would argue this actually was really helpful. Someone who is wrenching at home with not too much knowledge may be able to diagnose some of the issues they have using GPT.
You definitely are my favorite channel to watch !!!
Love the way you explain things in detail.
Stay blessed.
And good Outro Max..
Wow AI trying to diagnose engine issues, pretty wild stuff there! Glad you made this video for us to look at Wes view of the coming technology!
I tried ChatGPT at calculating the centripetal force acting on a person at the equator which is about half a pound and it got all the equations right but calculated the wrong answer five times in a row. One of the answers was many times the weight of the person so he'd fly into the air. I already knew the right answer from an online calculator so I just kept telling it to try again and it apologised and screwed it up again. It's very good at other stuff but it boggles the mind that the one thing a computer can't do is calculate.
That's because the way it works is just predicting the next word based on probability. If you use gpt4, it can do math because of a Wolfram alpha plugin.
The calculation cant work anyway because gravitational theory is no good. The earth cannot be spinning we would be in a centrifuge like environment where most light things are blown away from the surface of the earth.
@@JayDee-b5u Back under your rock troll.
@@JayDee-b5ulolol at gravity doesn't exist I guess😂😂
This is the only video on Chat GPT worth watching. But you sir make a great video. Thank you.
Really liked your approach to both the AI angle of this video, and your general everyday work. Thanks and all the best.
What do I think? I think that it’s a bright spot in the week when I see a video from Wes. Always entertaining while being informative. No way in hell any AI will replace an original like Wes!!!
Great video. Well presented content. Never disappointed with your channel. Thanks for taking the time.
Good overview. I suspect over time, this technology will mature or get bought up and destroyed. The internet (dumpster fire) is a great example. Great, diverse content.
I tried to prompt it to give me the best steps to diagnose this issue based on the codes to minimise time and money spent on diagnosis and it basically gave me similar steps. I think because replacing the Cat is expensive and time consuming so it had some sense of that as a last resort. Which isnt totally irrational. 🤷♀️
This is the output I got:
The combination of error codes you provided for your Chevy Trax can indicate a few potential issues. Here are the most likely causes and the steps you can take to diagnose and minimize the time and budget spent:
1. P0171 - System Too Lean: This code indicates that the engine is receiving too much air relative to the fuel, resulting in a lean air/fuel mixture. The most common causes for this code are a faulty Mass Air Flow (MAF) sensor, a vacuum leak, or a clogged fuel injector.
- Start by inspecting the air intake system for any leaks or loose connections. Check the intake hoses, clamps, and the air filter box.
- If there are no visible leaks, the next step is to clean or replace the MAF sensor. It's located between the air filter and the throttle body.
- If the issue persists, you may need to have the fuel injectors professionally cleaned or inspected for clogs or malfunctions.
2. P1101 - Intake Air Flow System Performance: This code typically indicates a problem with the Mass Air Flow (MAF) sensor or the Throttle Position Sensor (TPS).
- Since you already checked the MAF sensor while diagnosing P0171, verify that the sensor is properly connected and clean. If necessary, clean or replace the sensor.
- Additionally, inspect the wiring and connectors related to the MAF sensor and the TPS for any signs of damage or corrosion.
3. P0420 - Catalyst System Efficiency Below Threshold (Bank 1): This code suggests a problem with the catalytic converter or the oxygen (O2) sensors. The catalyst's role is to reduce emissions by converting harmful gases into less harmful ones.
- Before jumping to conclusions about a faulty catalytic converter, check the O2 sensors first. There are usually one or two sensors located before and after the catalytic converter.
- Inspect the wiring and connections for these O2 sensors, ensuring they are in good condition.
- If the sensors are functioning correctly, then a failing catalytic converter may be the issue. Consult a professional mechanic for further diagnosis and potential replacement.
By following these steps, you can cover the most common causes for the combination of error codes you mentioned. Remember to proceed with caution and consult a qualified mechanic if you're unsure about any steps or if the issue persists.
You so right about needing a confidence level. I’ve asked it to do that and it kinda worked for that one thread.
Also it just a matter of time before Chat gpt results are cluttered with ads and sponsored links.
You never fail to entertain and inform, thanks Wes. 👍
I've been working in IT for over 35 years now and lots of new tech has come and gone in that time. Some of it was great, some of it was garbage. AI is the first tech in all of that time that actually scares me. People are trampling all over each other to implement it while AI experts who have spent their entire careers working on this are pleading for people to stop. This isn't going to end well.
As for the Internet in general not all of it is a dumpster fire...there are still plenty of options out there in terms of software, browser add-ons, privacy-respecting sites, etc. that you can use that eliminates 95% of the garbage.
Would you care to list some of those options?
Wow. BS much? No one has been in AI their whole career - unless their career is so short as to disqualify their opinions. No-one is pleading for it to stop. Elon musk has made a few comments, but considering he is the dumbest smart guy out there, who cares? Anyone who thinks moving to a planet w/o an atmosphere (to speak of) is a viable plan has zero common sense. every accident will result in death death death. dumb as the clown going to 4,000 meters with a window rated at 1,000 meters. DUMB! this is the same guy who spent 44 billion on twitter which makes almost no money. DUMB! just so he can piss off the libs - who are the main potential customers of his electric cars. DUMB! Sam Alman? Where do I start? WHACK!
Thanks for questioning the hype! Since you had already found the issues it was interesting to see how AI approached the problem. Since we are stuck in the age of truthiness it's refreshing to get some factual feedback and a reality check! Great video!
P.C.V=Designed vacuum leak. ~information is always helpful no matter what source it was generated from, as long as you apply it correctly.
I think your last point is bang on. Google searches have gotten way outta touch with promotions... All of which are irrelevant to the search
Hey Wes keep up the good work I love watching your videos even the electronic billiard videos cuz it kind of helps me sometimes if I have a wiring problem I love your videos Man
Studied AI (well, neural networks really) in 1997/8 and from what I can see, nothing has changed about the tech in any major way, but now with the internet, it has access to untold amounts of data. Because it's machine-based, it can remember all that stuff. So it's basically a big rememberer
Right. I remember them talking about AI in the early 90s. Movies like Jurassic Park or the X files were all about it.
3 out of 3 on exhaust flange bolts?!? Wes, you're living right. Ya fortunate son-of-a-gun. 🤣
I think I used up all my luck.
Just wanted to chime in on my VERY narrow experience with ChatGPT. I use it every single day. As you said, it's fantastic with programming. Its ability to convert natural language to code and even integrate your own variables is fantastic. I've also learned some new PowerShell techniques. I come from a long line of scripting languages (20+ years of large-scale Identity Management and MS AD experience), so getting rid of older bad habits I've picked up from batch programming, to VBScripting, to Python, and now Powershell.
It's been great with most of the war history stuff I throw at it as well. I think your assessment is spot-on. Appreciate your multi-disciplined perspective. Like I mentioned, mine is super narrow. Cheers, brudda.
This was a cool video! Thanks, from a former VAG mechanic!
One of the few ChatGPT videos out there, other than Computerphile that points out it is a language model. Great job Wes.
Someone asked chat gpt what color blue was, and got some strange things for results, like "blue is a primary color", or "blue is a color in the visible spectrum", but it also said it didnt understand the question.
I agree with you on the AI issue, not right now. It was a big todo in the machine manufacturing industry too (machinist) however it has come closer in that arena than any other with CNC mills, lathes and grinders with automated pick and place equipment. Thanks again and, keep on smiling.
I worked in CNC manufacturing for years. People vastly underestimate how easy or even possible real automation is.
You hit the nail on the head with how bad internet searches have become. Kinda the end result of monopolies, I guess but still horribly annoying.
Brain, knowledge and new technology is the best combination for good car mechanic!
HA! The leaky valve gasket lubricated the exhaust flange bolts just enough to let you un-bolt them without surgery!
I just worked on one of these for a friend and I had all those same codes and it wound up being the PCV valve stuck open but since it's molded into the valve cover it wound up costing quite a bit more that it should have
Yeah same problem this one was screaming when it was running so no need to get out the smoke machine
I feel, like a shade tree mechanic (or a young person just learning) might find these suggestions useful. I have no doubt that every mechanic knows lots of things, but I don't believe there's a public-ally accessible database of those things, and I don't know if all the TSBs are out there either. I'm going to say, that based on other things ChatGPT has answered, those answers were actually pretty good.
Obviously you are cut above many machines, good job. I love your class room style
Totally agree with your assessment of ChatGPT. I was initially pretty freaked out by it but as I've learned more about its limitations I've become less and less impressed with it.
Yeah…..but have you tried the new WD-40 with AI??? It actually cusses at the frozen bolts for you. Pretty revolutionary.
I thought he'd broken his hatred of the stuff by using it on the manifold nuts but I guess it was just leaked oil since they were bone dry underneath.
Just discovered your channel a few days ago and have been binge watching.
Love it!
Also:
1995 Unicorn dodge. Change the steering shaft! Everyone overlooks it because it’s hard to determine how worn it is. Trust me, I am 100% positive the wondering is due to a worn steering shaft!
You asked for opinions from people who know…
Brother, I know!
Thanks for the videos!
I’m a repair guy and I use chat gpt all the time. It’s a powerful tool. You have to know how to roll the dice, and reroll. I ask the same thing four ways with different qualifiers. Super helpful for teasing out issues.
Good video btw.
Hey, Wes from the south here! Great job in explaining and showing everything. I do these valve covers all the time on euro cars. I agree with you about the AI doing diagnostics but it will never replace an experienced tech. I bet if the AI had all service information and TSB's it would work better rather than random web answers.
I’m so glad that you referenced that the fact that chat gpt is merely a indexed tool that simply goes through all the data it’s been fed and literally finds the most probable answer. Another great video with great explanation 👍
that ABSOLUTELY is NOT AT ALL how LLMs or ChatGPT works. not even a little bit. it is not intelligent or sentient, but your description is completely wrong. During training, the model is shown a sequence of words and taught to predict the next word. The model assigns a weighting to each part of the input data based on its significance and changes the weightings based on the difference between its prediction and what the next word actually is. This process of self-supervised learning is repeated until the model reaches the acceptable level of accuracy before an evaluation is carried out using a test dataset that was not used during training.
@@jasonbender2459Exactly, in other words - it is not retrieving full sentences/paragraphs like a search engine, it's generating unique sentences depending on how you ask the question.
For anyone who wants to see proof of this, try asking it a question that should modify the answer in a way that doesn't exist anywhere on the Internet, like 'Tell me the testing process for these codes, but from the perspective of a time traveller from the past who has never seen a car before'
When testing AIs they make sure the test data is NOT in the training data. That's called contamination.
Yet the best AIs are out scoring human experts.
Indexing? Not at all what's happening.
Excellent summary at the end, Wes. Keep in mind that very few people are as analytical as you are, so AI will (when used properly) assist the general population to think through problems and find solutions. But it is only in its infancy. It will almost certainly develop to a point in five or ten years to where we won't even remember its current limitations.
A general population thinking and deciding things for themselves is what they're most afraid of..
A couple years ago my 1994 F150 developed an issue that was an absolute pain to try and figure out. Took *months* of on-and-off diagnosis to figure it out, but I eventually did.
After watching this video, I decided to try this out myself, to see if ChatGPT could have saved me any time or if it really was as hard to diagnose as it felt. I fed the following symptoms into ChatGPT:
1. When started from cold (i.e. after it's been sitting for more than 8 hours) or hot (i.e. if it's been sitting for less than 20 minutes) it starts up right away. However, if I try to start it at any point between those two times, it cranks for an extended period before it starts. If I push the accelerator all the way to the floor it'll usually start up fairly quickly.
2. While it's warming up and the computer is in closed-loop mode, it runs just fine. But as soon as it enters open-loop mode, it starts misfiring like crazy at idle.
3. The misfire ONLY happens at idle. When I'm giving it throttle to either accelerate or maintain speed, it runs perfectly smoothly. It only starts to misfire once you come to a stop and the idle drops.
4. The misfire is very evident if you stand by the exhaust. You will periodically hear a "chuff" when a misfire occurs. It happens several times a second but the spacing between them is not consistent.
5. The exhaust smells of gasoline, which to me indicates that it's running incredibly rich
6. When I run a KOER (Key On Engine Running) test, I get diagnostic codes 173 (Lambda sensor too rich) and 70 (cylinder #7 failed cylinder balance test).
CGPT suggested, in order, that I check the fuel pump, filter, and injectors, the spark plugs and wires, the ignition coil, the distributor cap and rotor, and the O2 sensor.
My dad and I had tested and/or replaced all of these things (except the fuel pump, because it happened on both tanks so it was very unlikely that both pumps were failing in the same way at the same time).
I then opened up a new instance to clear out CGPT's memory and fed the same symptoms in, but omitted any mention of a KOER test and the DTCs. The *very first thing* it suggested I do was check to make sure that the fuel pressure regulator wasn't leaking fuel into the intake. That was what ultimately had fixed the truck in the first place, and it makes me feel better about how long it took to figure out. Goes to show that just because a car is computerized doesn't mean the computer will give you helpful information (particularly OBD-I. I've seen toasters smarter than early ECUs). I chased my tail for MONTHS because of those stupid trouble codes.
FWIW, that's a common problem. It can be a real head scratcher.
Best channel of this nature on the whole of YT, great work!
I agree with you wholly about AI. I prefer reality, not plastic, artificial crap.
Now I'm going to saddle a horse and check out a good trout stream to use my bamboo fly rod in .
How's that for reality ??
I like watching you figure out stuff and fixing it right .
I've been following technological trends since I read Ray Kurzweil's book Age of Spiritual machines around 2001 or so. I can say that we're not at the AGI stage yet, even with GPT 4, but the quality of this thing is quite the milestone that almost everyone said wouldn't happen within our lifetimes. As far as I can tell, we're a little behind schedule on Ray's predictions, but we're not too far off. I think he said we'd be able to buy a PC type device with human-level intelligence (or at least human brain computational capacity) by 2029. I think the trick they'll have to pull now is integrating things like Wolfram Alpha (which Steven Wolfram is actually working on) to fix all of the uncertainties that the LLM model entails. The next step will be miniaturizing and upping the efficiency of it so that it doesn't take up a whole warehouse to make one of these go. There's been a lot of work in the open source community on that front, to some success. Check out a fella named Brian Roemmele if you'd like to learn more.
I think it's really interesting that the things it gets wrong are the same kinds of things that people are generally bad at. Its memory is murky, it's not good at arithmetic, it lies instead of admits it doesn't know or it's not sure, but damned if it doesn't talk just like a person. These flaws are all very human-like qualities, which was the thing that startled me the most. We think of computers as having this indelible perfect memory but turns out that if you stack millions of linear functions together as if they're neurons it acts an awful lot like a stack of millions of biological neurons and becomes terrible at math lol. The emergence of ChatGPT tells me that AGI will be here within our lifetimes. I figure another 5-10 years due to the exponential nature of these things. Now that they know what makes it go, they just need to optimize it, so it's only a matter of time. The incentives to get there first are astronomical too. This could end up being Pandora's Box or the greatest thing we've ever done. Only time will tell.
I'm a software engineer with a substantial mechanical background and ChatGPT has become an almost integral tool for me, despite the countless complaints I have about it.
However I've gotta point out that (as you said yourself) it gave steps to follow (generic or not), it provided feedback on your responses (generic or not), and it did in fact lead you to numerous correct solutions.
Also I think that the "it doesn't know all the known problems for this car" is simply just something nervous people would say. There's many mechanical forums online that ChatGPT has access to, but the one thing it doesn't have is bias... Hence why only people say "X cars are chunks of crap". You also had the privilege of seeing the car in person, and I'm keen how you think you'd compare to ChatGPT if you were to be diagnosing the car over the phone?
For someone who's broke and not mechanically minded this could have potentially saved them hundreds of dollars at a dealership (if not more).
So for what was meant to be an impartial review it felt a bit like a "tooting ones own horn" session to me 🤷♂️
I'm as old as dirt. I hate confusers but we are forced to use them just to make phone calls, access our money, listen to music, or watch tv. I've said it before Wes, I'll stick with carbs and points distributors-and Briggs and Stratton flatheads. After over 80 years, they still do the job and can be fixed and maintained by us mortals. To bad we can't get good parts anymore. Thanks for posting.
I have a 2013 Buick encore same engine, I replace the exhaust flex pipe with a new GM part, had no trouble with nuts coming off. My check valve failed in the intake, so I used a plastic plug and JB weld epoxy to close the hole permanently and replaced the valve cover. I also replaced the flex pipe from the intake to the turbo and cleared the codes p0420 and p0171 on the car and the codes never came back and the engine runs smooth. In my opinion you should have change the timing chain tensioner between the Cam phasers when you had the valve cover off, it is very common for the bolts that hold it down to come loose, that's why you hear a rattle in the engine. I can hear the rattle when you start the engine at 17:40. I replaced the tensioner and clean out the thread holes and used lock tight for the bolts. Great video though, I like how you explained the PCV system on this car, I seen a lot of other video's on this engine and you are the first to explain it.
Yeah, that's one thing I noticed about using google or any other search engine for that matter. It's just a bunch of crap popped up to you, so it definitely felt good to have something like chat gpt help in that regard.
The long term fuel trims usually take a couple of trips-- and/or a certain amount of run time for it to relearn and walk back down to the correct readings. When you see higher long term (such as 15%) and at the same time you see a negative short term (such as - 3% or - 7%, etc). That is the computer telling the long term setting that it's too high and needs to walk back down-- which it will do over time. Some cars learn very quickly and others have certain triggers that you have to hit first; like being fully up to temp and doing a complete trip, etc.
Wes, you should ask ChatGPT how to keep mice out of your shop.
As a professional nerd, I've found that ChatGPT can sometimes be very helpful, saving lots of time searching for answers. However, sometimes it is totally wrong in its responses, and unless you know that, you can't call it out. I usually reply with "Are you sure?" with every response, and half the time it will correct itself.
Same here
Wes, a couple of things:
1. Prompting, or telling the AI what you want, is a whole skill in and of itself. If you give it generic "what do?" prompts, it will give you generic advice. I imagine you could have gotten a bit more useful advice had you told the AI your troubleshooting steps "Okay I went on a 20 minute test drive, here are the data points, I observed these, what are some likely causes?". You are right in that it doesn't have access to TSBs - but you can copy and paste the TSB into ChatGPT to give it some context as to what might be a good next step. It does have a good memory.
2. GPT-4 is an entirely different ballgame to GPT-3.5, which is what you used. It is hard to understate what a massive leap forward in intelligence and accuracy GPT-4 is. To put things in perspective. I don't know much Python, but I've told GPT-4 what I want, take that, run it, and iterate on that code going back and forth with GPT-4 until I have a functioning program. It is nothing short of miraculous. The only limitation is that it's 25 messages per 3 hours and it's slower than GPT-3.5, but it's great for complex problem solving.
I mean of course it won't steal your job, 90% of the work is physical, but the fact that it can help the layman diagnose the issue is really nice
could you google most of this? probably, but you're gonna have to visit at least 5 different ad-ridden websites for all the info, here it's compiled into a neat list
edit: I wrote that last sentence when I didn't fully finish the video, you actually addressed this exact issue, love it!
It seems like you had an interesting experience diagnosing and fixing a car issue with the help of Chat GPT. While AI can provide general information and suggestions, it's important to remember that it cannot physically inspect or diagnose a vehicle. However, based on the information you provided, Chat GPT suggested inspecting the intake system and mass airflow sensor, as well as checking for exhaust leaks before the catalytic converter. It also correctly mentioned that an exhaust leak downstream of the catalytic converter can affect oxygen sensor readings and trigger the P0420 code. You discovered an exhaust leak behind the catalytic converter and a vacuum leak in the PCV system, which you fixed by replacing the valve cover gasket and addressing the leak with a Quick Connect clip. You also replaced a faulty downstream O2 sensor. Regarding your thoughts on AI and its impact on jobs, it's an ongoing topic of discussion. While AI and automation can replace certain tasks and job roles, they can also create new opportunities and enhance productivity in various industries. It's important to adapt and upskill to remain relevant in a changing job market. AI, like Chat GPT, can be a useful tool in certain applications, but human expertise and judgment are still essential in many domains. Lastly, your observations about the state of the internet and the simplicity of receiving plain text answers from Chat GPT are valid. The internet can be overwhelming with ads, trolls, and information overload. AI models like Chat GPT can provide straightforward responses, but it's important to critically evaluate and verify the information obtained from any source, including AI models.
I like chatgpt...it gives detailed explanation and solutions. However, it doesn't have information after 2021.
Great video 👍👍
Good job Wes! I think that you killed it on this job!
It managed to diagnose my 2005 mini cooper s r53's error code P0130 02 sensor issue
AI's prevalence and growing influence on our daily lives can indeed feel overwhelming, but it's important to remember that the progression and adaptation of technology has been a part of human society for centuries. While the rapid pace and complexity of AI can seem daunting, it also offers a variety of opportunities for innovation and improvements in efficiency. Just as past technological revolutions have reshaped industries and lifestyles, AI is poised to do the same. As with any technology, it's essential to be aware of and engage with the societal implications, which include job displacement, privacy concerns, and ethical dilemmas. Furthermore, a shift towards AI doesn't mean that everything becomes automated and human input becomes obsolete. There will always be areas where the human touch is crucial, especially in roles that involve creativity, critical thinking, and emotional intelligence. If the constant talk about AI is tiring, it may help to focus on learning about areas that you find most intriguing or relevant. For example, if you're a healthcare professional, you might want to learn about how AI is used in medical diagnosis. Or, if you're an artist, you might be interested in how AI is being used to create new forms of art. Moreover, public participation in AI discussions is vital. These conversations are not just for tech experts; they affect everyone, so everyone's voice is important. Remember that the goal is not just about 'surviving' the AI revolution, but shaping it to ensure it benefits all of society.
This is exactly the type of research we need if we want to do more high end work!
You are spot on with your opinion on the internet,,, but wait chatGPT will default to the same clutter..
I use chatgpt for my programming projects and general questions. I know its answers arnt always correct but i like that i dont ever need to LOOK for the answer it just gives it me. So even though the answer it gives can sometimes be wrong overall it usually saves me a ton of time.
Super interesting and never done before by any car channel 👌🏽
Very interesting evaluation. However, what was more enlightening for a former shade tree hack, is all the knowledge and equipment needed for just one manufacture's model and the Rube Goldberg plumbing required to make these things run with minimal pollution. Anyone who complains about the cost of repair would do well to watch this video.
I luv your vids, you r right about the net, I always learn something about Auto Tech when I watch, Thanks Wes