I don’t know how ai works nor can i fathom how but if you manage to reward consistently more at the start of training and then pacing afterwards you might get a more efficient learning curve

very facinating! Extremely thorough and good resarch. thumbs up from me!

When is the next chapter coming?

What an incredible explanation of how an AI works, how it is trained and the GREAT example of what our universe is, CHAOS. It is one of the best videos I have ever seen in my life, my sincere congratulations. you deserve more.

What an amazing video!!!

dude, make more videos beating WRs!!

Great, now it's time to create an AI which don't need hours to learn any track just not to crash. Still far from human intelligence, but for sure it will significantly improve.

I think, even if physics is deterministic, the framerate is not. The calculations happen with some Δt between frames, and Δt is affected by the multitasking nature of the environment the game is running in. There are I/O interrupts, reschedules to different cores…

L'accent français dur

I’m an Econ major getting a minor in complex systems and this definitely is an application of chaos theory. I think the moment generated off of the jumps is just so sensitive to the cars initial conditions that possible even rounding variations in certain variables could introduce randomness and variation in attempts.

AI spamming the drift buttons is too funny… it discovered drugs 😮

Do you have the model code.

Imagine a very optimized code running on a super computer assisted by a multiple quantum clusters. In a few days. The times would be impossible to beat at all.

It is like normal life, someone imbeded in our life's reward/penalty system (and other) to figure out a meaning of our life's 🙂. It's preety good way to finally understand that God programed our Universe and WE have full life (like this truck) to find best way to our destination 😀. We have that advantage that we can learn from each other so it's faster. So, as much as AI is getting better, but it is dueto YOUR input, adjustments etc. it's getting better. It's not getting better on its own. It's not adding parameters that would improve its own progress. Ps. Can AI simply finish race without falling of the edge on a first attempt without being the fastest?

Tu es fr ?

You should take this ai and put it on a random map. The only time it gets a reward is when it beats its old time. Let him train for 2 hours. Then you train for two hours. Then yall race

are you french?

6:50 every one failed 😂

OH MY PCCCC

0:06 ready player one ahh AI

SLOVAKIA MENTIONED

24:39 bro is just roasting humans by this point, but nice vid

quantum computer will find local minimums, it will be really good at chaos i wish I could see it happen in the future

oh yeah 272,000th like 🙏

'we got the AI working on a simple racing game, and it only took 3 years!' '...okay. and it can consistently handle all of the maps we draw?' 'no! only one! but it does SO WELL that it is no fun to drive against!' '...'

5:52 what song is this?

You should add a thing to know how far the next turn is to the next next turn so it can slow down ready if it's got two turns in a tight time or know to just try and go full speed

This is the most amazing AI project I've ever think. Even anyone without any clue about ML or AI could understand with the help of this video

Love how smart the AI is. Like you taught it something completely new in drifting and then it just did it on its own. Very human like

Braaaaaaaaaaaaaaaaake

There is ABSOLUTELY no randomness; how do you think replay playback works? It just does the same inputs. Note: landing bugs are still bugs, even if they aren’t random; what we see as random is truly just something that relies on factors we don’t consider reasonable, but that is not randomness, it is just the precise circumstances of the situation, which are equally “random” for practical reasons.

9:46 hold on let me go 78 stories down the stairs bc my jaw just fucking DROPPED

4:13 I would try rewarding it based solely on its finishing time at this point; obviously this only works once it learns how to consistently make it to the finish though

Un français qui a bien percé en anglais😊...

can the ai learn from your gameplay? I think if it can it would be much more faster

Im youre frend now :)

9:30 So was it the fact that the ai's back tire would catch a corner, allowing it to take a tighter turn?

you should add a thing where if the ai hits the wall above a specific speed it looses all rewards

May be your computer can’t run everything all at once causing a slight change in starting time

Heyy, when is the next video coming out?

"OH MY PC" 🗣️🔥

Wow, this is fascinating! My son is currently learning game development on the Moonpreneur platform, where he’s immersed in cutting-edge tech and learning how to push boundaries. Seeing AI excel in challenging environments like Trackmania, breaking world records, really reinforces the importance of innovation and adaptability in today’s tech landscape. It’s inspiring to see how AI continues to evolve and set new benchmarks!

when you mastered a game and try to make an AI to beat yourself:

Negative reward for touching wall and rapid deaccelration

@Wirtual

I like how he pronounces “knowledge” like “noleidge”. It's so funny.

Toi tes français 😂

when will be the next video :))?

Is the AI blind?

Pas mal l’accent francais incroyable la video