I create videos of AIs doing whatever crazy shenanigans I can dream up. It'll mostly go horrifically wrong and they'll find new wonderful ways of disappointing me, but occasionally impress me and persuade me not turn turn off their life support for another 5 minutes. I enjoy everything AI related, and long walks on AI generated beaches.
One day we will we able to swap cpus with ais and the gaming world will be never the same. Imagine playing you favourite game against a learning ai that would be so fun
I thought i'd let you know if you're making an ai that plays online, theres a decent if not very good chance it will be detected. I know what I'm describing is not your field, but there are controller mods for games like call of duty where it overrides player input to make better moves, and cheaters fly under the radar by lowering the settings as to not be obviously too good at the game. I wish you the best of luck, as its very unfortunate the bad eggs poison the water you want to work and play in.
I know you've made the ai play on multiple levels, but I'm curious if its more efficient to have a different ai trained for each level, and load one thats specific to that course, as opposed to have one ai model learn all the levels. I'm guessing the downsides is having to duplicate the model's file size for each case use on the hard drive.
25:50 - i can confirm. I found a lot of values in super mario kart, but the value for baloons in battle mode elude me. At best i can find the value for being hit.
I'd love to see how much this AI could learn on it's own given a bit more time! Armagedon is a game with an incredibly high skill celing, and I'd love to see how long it takes for it to learn some basics like special moves and parries, all the way to things like infinites
Hopefully you'll answer even tho I'm a month late lol. But whats the reasoning behind not first teaching the AI how to complete the track through human playthrough, then apply the deep learning once it knows how to finish. I'm aware that this would involve mixing multiple AI algorithms, but I feel like it would be worth it. You wouldn't have to wait several days until the AI can finally learn how to complete the track itself. The AI could use that extra time learning how to get faster times
for a future mario kart AI, you should have it play 8 delux. then, once its proficient against bots, put it onto the multiplayer scene! Also, just as an idea for the future, you should have the AI learn how to complete the tracks quickly from time trials where there are no other bots before dropping it into games with other AI's. Having an AI which can already drive around every track at blinding speed seems like a great place to start before putting them up against anything else
Hey there, do you think it is possible to use a trained mario kart ai and let it use TAS inputs as a foundation, give it like 40 hours to improve that? I mean even if it is like only 50ms faster after that, it improved a TAS and that would be kinda revolutionary for speedruns
This is an awesome series. Just found your channel and subscribed. Can’t wait to see what you have next! One idea I have to make the AI stronger, potentially by a significant margin, is to completely ditch Q-learning and instead switch to something similar to what DeepMind did 6-7 years ago with AlphaZero (or MuZero for a more recent iteration). Essentially keep training it in 2-player mode, and make it a champion/challenger model with the objective of the challenger beating the champion, and when it does consistently beat the champion, the challenger becomes the new champion. Rinse and repeat. I think one of the benefits to that architecture is that it forces the model to explore different ideas to try and beat the champion, and it’s much less likely to develop blind spots. There would be 2 major downsides with this though: (1) it would take significantly longer to train, and (2) the items boxes add a considerable amount of variability into the results, so you kind of have to run it 10-20 races (maybe more) for each training step to know if the changes made a difference. But I suspect it would figure out how to dodge blue shells, and potentially even figure out how to take expert shortcuts on certain tracks.
Ai tango hey you ever figure out how to play Mario kart 7 with ai I know the online servers are done but did you ever figure it out or you forgot about it? 😅
I like how when it goes for the trick jump, it hits the middle point of the choice between the left turn and right, then decides right based on that reward difference. It's really robotic and shows that it hits a very specific spot where it knows to turn for maximum reward based on the direction and turn reward of it versus the fastest turn being heading straight for it from that angle. Since it has to consider all directions when it's gaining the reward for optimal efficiency due to the "predicting" reward system it follows. I most noticed it in the Final Ai segment on the first laps.