Just thinking out loud here, but if you’re +2secs off the pace on a fixed session you should improve your driving technique first then start working on setups, IMO that helps to avoid bad habits as well, anyway good video thanks for sharing!
More than a year ago I did this in ACC. I had problems with tyre heating, and I changed the dampers as AI suggested, and it worked. BUT it was a hit and miss in some areas, even basics. You still need some knowledge.
The prompt engineering of these will be important. I think creating a custom gpt with loads of data for the model to reference will be the key at AI created setups. This is similar to what race teams do back at their headquarters when running simulations. Now the next step will be adding your telemetry data and the setup information for the models to recommend changes in setup or changes to your racing craft.
Interesting.I actually have tried to use AI to give me advice for setups in historical sim racing.Sometimes it's limited by lack of knowledge of those setups from those days.But sometimes it gives me some interesting insights😊
I would suggest to fine tune-it by adding some knowledge base of technical literature about setting up racing car chassis and aerodynamics. Surprisingly it did give me some very decent results after that. At least gpt was pointing the right things to adjust with better accuracy and in which direction to go. :)
But the critical thing is - What did you input into Chat GPT? What information did you provide, as in did you input the entire base line setup and then ask for changes?
I should've been more clear in the video about this but I wrote this as a prompt: I am driving the Dallara P217 in iRacing on the nurburgring GP circuit on the baseline setup Could you recommend some setup changes that add: - More precision in high speed corners - More traction during acceleration - More stability in high speed corners
Oh that’s awesome. That’s pretty straight forward, kind of like coming back to the garage and telling the engineer what you need from the car. I thought you may have needed to first outline all of the current settings. Love this, I’m going to try it 👊🏽
What if you asked AI to give you the best race setup "for" something, like long run tire wear, short run speed, stability, etc, and specifically for the car rather than suggestions to improve what you're struggling with? So, something like "IR18 race setup for stability at Long Beach Street Circuit on iRacing," or however you'd write it. Then adjust from there with what you're struggling with.
The driving was way too inconsistent for any meaningful conclusion to be drawn, perhaps if the laptimes weren't changing by 4 tenths every lap and you did 10 or 15 laps you could compare.
Ik ben geen simracer maar heb sterk het gevoel dat de setup van chatgpt niet perse beter is, maar je lijnen waren wel een stuk smoother. Misschien als fair test zou het beter zijn om de chatgpt setups te proberen als je weer een tijdje niet je sim hebt aangeraakt ipv dat je er helemaal in zit. En zoals andere al suggereerde in de comments, ik denk dat chatgpt qua kennis die het tot zich heeft heus wel goede setups kan maken, maar dan nog is het punt wel dat de coureur zn drivingstyle wel overeen moet komen met de setup die je krijgt.
So many missed apexes that the setup might not make that much of a difference. But still an interesting idea to use ChatGPT for this. Wondering if we people will start relying on ai more often instead of learning cool stuff deeper.
Honestly not being on the limit of all four tires while driving the correct line is a lot slower than the optimal slip angle through a sloppy route. To the merit of the video, changing the setup can make the balance a bit more neutral and responsive, the default tends to be really tight and takes a lot of precision to get the slide right, the overall grip may be lower but being able to drive it consistently is what matters the most. That being said, picking a slower car that spends more time on the edge would have more prominently highlighted the changes.
Insteresting video. The idea is cool, I wish you would have shown what you changed in the setup sheet, video should have been a lot longer. If you are about to redo a video with a similar idea, I would suggest you to get very familiar with the track, just take the track of the week from the series you are mainly driving, after a few races and using a real setup from a shop or what ever, you are probably better prepared for this challenge, using the fixed setup and not the baseline ( or you can use any baseline setup that are labeled with high, low, medium downforce ), since the purpose of the baseline setup is just to not fail under any load of fuel, there is no work on the set. You did 10 laps, 5 laps on each set.... low effort video or lack of knowledge, I don't know. Finally, can AI create a faster setup? - The driving and the lack of visual information for the viewers are the variables that make this completely useless.
Most of the time this will happen. A driver with, most likely, not e-sports levels of skill, with most likely a poor understanding of vehicle dynamics, asks an AI, known for spouting utter nonsense what to do. Results will be random, which means sometimes they seem right, which is not to be confused with correlation!
every patch and every update changes the physics. And ChatGPT is fishing the internet for information from players from 2015 ectect... Pointless! There are only very vague and rudimentary recommendations.
It may not have made him much faster, but it definitely looks more stable and consistent. That, for less experienced drivers, is worth tooling around with in my opinion. Top split in the LMP2 is running 1:46-1:47 from what I've seen.
@@RobinHood-yk8og Could be, but .4 sec average difference across only 5 laps on each set is a pretty clear indication that the set was beneficial, even if the set isn't responsible for all of the time gained. I would think the driver would have to be pretty familiar with the track to use it for this test.
@@FifthPinned Not could be. Is. You cant tell anything from what you've just watched about the benefits of the changes made. From the long list of proposed changes, we didnt see what was actually different between the 2 runs. Nothing is "pretty clear". It's all rather muddy. If the driver was pretty familiar with the track, you'd expect a better time. First driver I pulled off G61 with a 1:49.0, had 3 complete laps and a similar iRating to PowerQ. So with PowerQ's assumed familiarity with the track... and still posting 1:49's... maybe you'd agree that his setup is still irrelevant to everything you just watched. And that for the forseeable future any improvements are all going to be down to seat time.
![COLOMBIA vs. CHILE [4-0] | RESUMEN | ELIMINATORIAS SUDAMERICANAS | FECHA 10](http://i.ytimg.com/vi/TDjN4W3yy1k/mqdefault.jpg)
I'm not a fast driver, I know I make a lot of mistakes and in this video I'm just having fun with AI to see if it can create a decent setup!
Just thinking out loud here, but if you’re +2secs off the pace on a fixed session you should improve your driving technique first then start working on setups, IMO that helps to avoid bad habits as well, anyway good video thanks for sharing!
More than a year ago I did this in ACC. I had problems with tyre heating, and I changed the dampers as AI suggested, and it worked. BUT it was a hit and miss in some areas, even basics. You still need some knowledge.
Very true, you can't blindly trust the AI, but it can help you get to a decent starting point
The prompt engineering of these will be important. I think creating a custom gpt with loads of data for the model to reference will be the key at AI created setups. This is similar to what race teams do back at their headquarters when running simulations. Now the next step will be adding your telemetry data and the setup information for the models to recommend changes in setup or changes to your racing craft.
Interesting.I actually have tried to use AI to give me advice for setups in historical sim racing.Sometimes it's limited by lack of knowledge of those setups from those days.But sometimes it gives me some interesting insights😊
What happens if you do a generational build? continuing going back to the AI with more questions could generate a good set up by the end.
I would suggest to fine tune-it by adding some knowledge base of technical literature about setting up racing car chassis and aerodynamics. Surprisingly it did give me some very decent results after that. At least gpt was pointing the right things to adjust with better accuracy and in which direction to go. :)
But the critical thing is - What did you input into Chat GPT? What information did you provide, as in did you input the entire base line setup and then ask for changes?
I should've been more clear in the video about this but I wrote this as a prompt:
I am driving the Dallara P217 in iRacing on the nurburgring GP circuit on the baseline setup
Could you recommend some setup changes that add:
- More precision in high speed corners
- More traction during acceleration
- More stability in high speed corners
Oh that’s awesome. That’s pretty straight forward, kind of like coming back to the garage and telling the engineer what you need from the car.
I thought you may have needed to first outline all of the current settings.
Love this, I’m going to try it 👊🏽
I might start doing this because I suck at setups
It is helpful if you don't know where to start
Great Video! I’m planning to do a similar video but doing it in ACC. I will give credit to you for this fantastic idea. Thank you 🤙
What if you asked AI to give you the best race setup "for" something, like long run tire wear, short run speed, stability, etc, and specifically for the car rather than suggestions to improve what you're struggling with?
So, something like "IR18 race setup for stability at Long Beach Street Circuit on iRacing," or however you'd write it. Then adjust from there with what you're struggling with.
The driving was way too inconsistent for any meaningful conclusion to be drawn, perhaps if the laptimes weren't changing by 4 tenths every lap and you did 10 or 15 laps you could compare.
Never mind setups. Where’s AI crew chief at?
that would definitely be useful when asking it questions
Crew Chief is a third party app for iRacing
Ik ben geen simracer maar heb sterk het gevoel dat de setup van chatgpt niet perse beter is, maar je lijnen waren wel een stuk smoother.
Misschien als fair test zou het beter zijn om de chatgpt setups te proberen als je weer een tijdje niet je sim hebt aangeraakt ipv dat je er helemaal in zit.
En zoals andere al suggereerde in de comments, ik denk dat chatgpt qua kennis die het tot zich heeft heus wel goede setups kan maken, maar dan nog is het punt wel dat de coureur zn drivingstyle wel overeen moet komen met de setup die je krijgt.
Great video. I may start doing this too
So many missed apexes that the setup might not make that much of a difference. But still an interesting idea to use ChatGPT for this. Wondering if we people will start relying on ai more often instead of learning cool stuff deeper.
Yeah I should improve my driving before looking for gains via setups
Honestly not being on the limit of all four tires while driving the correct line is a lot slower than the optimal slip angle through a sloppy route. To the merit of the video, changing the setup can make the balance a bit more neutral and responsive, the default tends to be really tight and takes a lot of precision to get the slide right, the overall grip may be lower but being able to drive it consistently is what matters the most. That being said, picking a slower car that spends more time on the edge would have more prominently highlighted the changes.
Insteresting video. The idea is cool, I wish you would have shown what you changed in the setup sheet, video should have been a lot longer. If you are about to redo a video with a similar idea, I would suggest you to get very familiar with the track, just take the track of the week from the series you are mainly driving, after a few races and using a real setup from a shop or what ever, you are probably better prepared for this challenge, using the fixed setup and not the baseline ( or you can use any baseline setup that are labeled with high, low, medium downforce ), since the purpose of the baseline setup is just to not fail under any load of fuel, there is no work on the set.
You did 10 laps, 5 laps on each set.... low effort video or lack of knowledge, I don't know.
Finally, can AI create a faster setup? - The driving and the lack of visual information for the viewers are the variables that make this completely useless.
Where is that chat GPT ?
Most of the time this will happen. A driver with, most likely, not e-sports levels of skill, with most likely a poor understanding of vehicle dynamics, asks an AI, known for spouting utter nonsense what to do. Results will be random, which means sometimes they seem right, which is not to be confused with correlation!
every patch and every update changes the physics. And ChatGPT is fishing the internet for information from players from 2015 ectect... Pointless! There are only very vague and rudimentary recommendations.
The newest ChatGPT is real time data
@@TracksideViews yes real time searching the Internet. Also Informations from 2015
Is this the GP layout? 6 seconds off the pace, your setup is irrelevant.
It may not have made him much faster, but it definitely looks more stable and consistent. That, for less experienced drivers, is worth tooling around with in my opinion. Top split in the LMP2 is running 1:46-1:47 from what I've seen.
@@FifthPinned That's time on track, not the setup.
@@RobinHood-yk8og Could be, but .4 sec average difference across only 5 laps on each set is a pretty clear indication that the set was beneficial, even if the set isn't responsible for all of the time gained. I would think the driver would have to be pretty familiar with the track to use it for this test.
@@FifthPinned Not could be. Is.
You cant tell anything from what you've just watched about the benefits of the changes made. From the long list of proposed changes, we didnt see what was actually different between the 2 runs. Nothing is "pretty clear". It's all rather muddy.
If the driver was pretty familiar with the track, you'd expect a better time. First driver I pulled off G61 with a 1:49.0, had 3 complete laps and a similar iRating to PowerQ.
So with PowerQ's assumed familiarity with the track... and still posting 1:49's... maybe you'd agree that his setup is still irrelevant to everything you just watched. And that for the forseeable future any improvements are all going to be down to seat time.
Who pissed in your cereal? What is the point of leaving this comment ? You are just being toxic.