Background:
My oldest son is a leader in A.I. and presents at I.T. conferences. He used to do all his own coding. He has done zero coding himself since this past January. A.I. has taken over the I.T. world. He is leading the charge for his software company to utilize A.I. Now he speaks with Claude AI by Anthropic, and it writes the code for him. Then he just has to review it and tell Claude to "do it over". Claude does not even mind or get tired LOL. A.I. is not perfect yet, but it is learning, and getting way better, and doing sovery quickly. He showed me how it works. We had a conversation with Claude for 10 min and it wrote 1500 lines of code for him in an hour, that would have taken him 12 hours. Cost to his account: $10.72. Anthropic (Claude) is the current A.I. leader. My son showed me how to use it. Just ask Claude questions, and be as specific as possible. Honestly it was like talking to a person, or an excellent researcher.

Car Design:
I compared Claude AI and Google AI this morning, and asked them both the same questions. I started a free Claude account. I asked it questions about engine design, camshaft spec selection, gearing, and intended car use for 10-15min. (I hit my message limit so now have to wait 3 hours to continue using it for free.) So I then asked Google AI the same questions. I had very low expectations... but...

Result:
Claude was the big winner by a huge margin. It was actually way better than talking to a generic phone tech at one of the big cam companies. Claude's answers were more correct. I have gotten some off the wall opinions from phone techs. I would say it was on a par with Mike Jones or one the head guys at Bullet cams, but I even got much more detailed feedback this way. Claude did not hold anything back. It kept asking me questions until it thought it had enough guidelines from me about my requested setup. Only then did it offer a suggestion. Very cool. Scary smart.

Google AI parroted back to me the typical generic responses you would get almost anywhere on the internet. Not bad at all. But also not very specific. It pretty quickly referred me to several known vendors. Big deal.

But Claude was much, much better. I was basically blown away. "We" had a very in-depth discussion, and it came back with some very detailed answers, always comparing the Pros and Cons, and always suggesting alternatives. We got into cam duration, lobe center selection, gearing, hydraulic vs roller, vs tight lash hybrid setups, cruising rpm vs cam duration, cam duration vs gas mileage vs cruise rpm, and on and on. It was a very good "conversation" with excellent pros and cons. The information it kicked back in 30 seconds took me decades to learn. And I still learned a few things.

Give "Claude" a try!

It's free.

 

I am an old man and this technology fascinates me. When I do a search I find myself going to the SEARCH ASSIST in my DuckDuckGo searches more often. I am getting answers to questions that I already know, just to verify what info I am getting back is correct. I haven't ventured deep into CLAUDE yet, just a bit and it truly is mazing. Unfortunately it won't be long before the bad guys get ahold of this and I doubt that even the smartest ones of us will be able to tell a true story or picture from an AI generated one. I think I will ask my new friend CLAUDE how to tell the difference. Great thread idea here.

 

leigh that was a very interesting/informative post. As a baby boomer myself, I am going kicking and screaming into the world of AI. I find it fascinating and frightening at the same time. I always say please and thank you when dealing with AI so at least they will kill me last (LOL).

 

I have several projects both at home and at work that Claude,AI has made a significant difference in time saved. I have a library of about 200 powershell scripts that I have been using for almost 20 years. Claude went through them and made improvements on 192 of them, many were significant improvements. I had a problem at work where Claude also found the error in a pharmacy script application I was troubleshooting with the vendor for 3 weeks. I uploaded a packet capture file and all the error log files and it identified the actual problem in less than 5 seconds. The vendor corrected his code and that was the end of it. Claude helped me design several circuits and fix a few others. But it does concern me. If this technology ever has the ability to become self aware, we might be in trouble.

Closer to the OP question in engine builds. I ran a few scenarios with cam changes, head changes and exhaust combinations using Microsoft Copilot and it has some very interesting results. Take for example an L82 stock engine, then ask it to run through various available cam and cyl head choices and see what comes back. Give it part numbers and it will find the specs, run the calculations and give results. I tried a few times to trip it up, like suggesting a replacement head that was for a BBC but it caught the error. Ask it to consider piston or rod combinations and it will calculate all that and give you a result. Id like to see someone compare an already built combination with known dyno numbers against the AI's calculations to see how real world stacks up.
I had an idea how the swapping to LED's in an instrument cluster could also be made dimmable by using the stock headlight switch. I wrote up my scenario and Copilot confirmed not only that it works but why it works. I will post that in another thread for those who prefer to use LED's in the inst and gauge cluster. I tested it, it does work
When it comes to coding, scripting, electronics etc Claude AI is ahead of the rest, Although I find Copilot to be about the same when it comes to scripting, but not as good as Claude with coding. Copilot is much better at project management, Chat gpt tends to blabber too much in my experience and the Amazon AI Rufus is just an absolute idiot. But they got deep pockets so it will probably get better pretty soon.

 

I just continued my Claude discussion, and it accurately predicted my actual dyno numbers from when I had it dyno'd.

WOW!

Was I impressed!

 

Thats what I was wondering about. How does a known build with actual dyno numbers compare to the calculations of an AI. Im not surprised by your results, but it is cool

 

Interesting read... but what struck me is that, I am thinking, in order to get quality out of Claude, et. al., you have to have the knowledge and experience to ask the correct questions and set accurate parameters. You have to also be able to evaluate the results to determine feasibility of the solution.

This reminds me of the criticism of kids using calculators in school and how it's a crutch and hindrance to learning math. However in order to use a calculator you have to understand math to identify the parameters and needed outcome, enter the proper figures and functions in the proper order, and evaluate the result for reasonableness.

Without a proficient understanding of the processes, whether solving a word problem or designing an engine or writing code, the output is unreliable and potentially costly. AI still needs educated and skilled humans at both ends. Leigh and his son possess that level of education; if I were to ask AI for engine design or software code the outcome would be far different and likely insufficiently complete.

 

Yes my ongoing engine discussion with Claude is now up to 88 pages!
And most often it is very good.
But once in a while it says something stupid, and I have to correct it.

It is just like the smartest teenagers in the school that were in my AP Chemistry class.
Very good,... but ... not always correct.
Usually they both just skip over something that experience says is important.
Just challenge it and it will correct itself.

But I had fun testing it to it's limits.
I currently have a 500 HP 550 TQ 454. Not bad. That fits under a BB hood, no L88 hood.

"We" kicked a lot of parts & changes around, and I now understand much better how to push it's "normal" 13mpg to 20mpg or so, and how to push it up to 670HP/670TQ even while keeping the same cam and keeping the same BB hood.
I thought that was cool, ...and worth my time.

It keeps drawing on hundreds of engine builds that are out there on the internet, and only applies what is important.
I can do the math, but it does it so much more quickly.
It even calculated thermal expansion rates, on my engine parts, and predicted the change in my valve lash from cold to hot, especially important on a .006" lash cam.
It gave me Dyno graphs, optimum ignition timing MAPS (far beyond a simple distributor curve), and target AFR readings and EGT readings.
All for my specific engine combo, and based on math, not just some "average" thumb-rule result.
Cool stuff.

I might even test some of it for real!

I guess I will leave the extra 170HP "on the table"....for now.

 

Yes. I absolutely agree with this. Its a great intellect, but has NO life experience.

 

I’ve been using Claude at work to do a lot of coding and it is amazing. On my 1981 Corvette I’ve been using Grok. It is fascinating and has helped me tremendously. I swapped out an old engine in the 81 and put in a new blueprint and could not have done it without Grok helping me. If there’s something I can’t identify I take a picture in Grok tells me what it is and whether I need to keep it with my new engine or dispose of it. It’s helped me on every facet of the 1981 restoration.