Oh hey tena, remember me ? ;D

lol

@@ceodasala3297 tough

They should add the Nexto bot to be Unstoppable as dofficulty setting bot in offline Rocket League.

been called that for years, new around here?

Most underrated comment.

So true

“Detroit become rocket league” is a wild comment and I love it! Lmao

“ “Detroit become rocket league” is a wild comment and I love it! Lmao “ - 🤖

@@noahgrimmer5095 robot ahh response

@@weymoo oh fuck their onto me

Fire comment

😂😂

@readmyname. we dont really care mate

@@tomaskoza4439 mate that account is a bot lol

you can just look at the spectate and who's spectating. it was to easy for them to guess

The companies are just going to have to wait. For countless companies players have coding their own versions of stuff they should of added for years. Some players even said they would give it to the company for free but these selfish companies never accept it as long as people are buying there products. If this becomes an issue and rl devs cannot figure it out then a player should.

@@averagebrownie2491 those evils at psyonic can suck on my cock hope that this game gets flooded with bots serves them right

As someone who plays TF2 as well I'm worried for Rocket League's future

I'm definitely not an expert but I think it's way easier to train an ai to recognize wether a player is actually a human or if it's a bot than to actually train an ai to play rocket league. If even humans can easily differentiate between ai and player that shouldn't be too hard, it's basically impossible to train an ai to actually play like a human. So my guess is you have a programm/ai watching the games and filtering out players with suspicious/bot-like behaviour, then you can even look at the accounts match history etc. and automatically ban it if theres no doubt about it. I have no idea wether that is easy to implement tho.

honestly once ue5 rocket league comes out it will delay the bots because it will have all the coding completely rebuilt from the ground up

I'm actually surprised (and relieved) that the exploit users didn't try to implement the other ML bots, like Necto or Element, so as to throw them off the scent. Maybe they thought they'd still be too similar, so detection would still be easy?

@@TrulyLegitGaming bro can’t even get gc rewards on his own lmao

@@TrulyLegitGaming gcs will easily tell a bot apart from a player lmao

@ToxpoV well so far they haven't been able to tell. And I've played well over 50 games with the bot

@@TrulyLegitGaming they've probably been able to tell, they just dont say anything. Also... you are cheating for gc? Gc isnt even hard anymore.

Tru tru

Usually bots tend to have "an exploit", the more exposure this bot gets the quicker one might be found, and depending on how easily exploitable it is it can render it practically useless at a certain rank. (I don't know too much about nexto, but let's say it wouldn't challenge if a player is dribbling at a certain angle, then a player who is decent at dribbling can use that to consistently beat it and so on)

I think everyone is focused too much on the BOT instead of how the BOT got into ranked in the first place. This is the real issue. We've already seen hacks (demo chase) and people using non-purchased skins in ranked.

@@VilTheVillain the trick is demos. Bots don't have the awareness of a human player and/or aren't programmed for demo evasion, its a major flaw in nexto's game. most players that play slow and controlled are susceptible to a high speed drive challenge, you either demo them or make them jump/flick and you're already zooming on the ground so you'll recover much faster

pros Will not use nexto in pro play, I am gc2 and bots are very bad compared to high level players, and I’m speaking from experience. I had a bot in my gc2 2v2 game and he was not very good lol

@@solarft give it time. Eventually they will be better than pros without a doubt. They havent been out very long

Haha yes and I'm gonna name myself "nice flick" to troll people XD

nah youd know.

Sometimes I think its some low level bot, when they all have the same name and end with 1, 2 or 3 and they move in very linear way, like their programmed to moce in that way.

Bro why is this scary to you. It’s not like they in your crib ready to hang.

bro why the fuck is it scary? it's a fucking video game bro

Bro, its already a thing watcha talking about xD

Actually a good idea. Imagine using traces of detecting bot-like behavior, and then randomly pairing the suspected account with a bot-detecting bot that is designed to analyze and perform actions that expose the account.

Just some notes I'm copy-pasting to address common misconceptions: Point #1: "Detecting AI with AI" - - - - - As someone who has had experience with developing neural networks and working with machine learning, people who say that training an AI to recognize bots, are wrong. Making an AI to detect bots is NOT easier than training the bots themselves. Generation always trumps detection. This is because of a phenomenon I like to call Detection Singularity: For example, lets pretend an AI has been made to detect AI-generated melodies. Eventually AI will have gotten good enough to generate great-sounding, catchy melodies. But, what happens if, by chance, the AI generates the melody to Mary Had a Little Lamb? It's a pretty simple, catchy melody, and an AI could easily generate it. But, should the detector believe it to be made by AI, or made by a human? When an AI melody generator has gotten good enough to generate human-made melodies by chance, this is called the Detection Singularity. It no longer becomes possible to tell whether a melody was made by AI, or by a human. Now, melodies are pretty simple, as compared to gameplay, which has many more variables, but there's only a finite level of complexity, and AI improves at tasks exponentially. At a certain level, even a game as complex as RL will reach the Detection Singularity, and it won't take long for many games to fall to this. Nexto was made with no funding, and did not even use heavy-duty cloud computing, which is becoming more and more affordable by the minute. Point #2: "The AI would become perfect, as they are designed to get good at the game, not to seem human. Making it seem human is much harder," - - - - - Some people may say that AI is more likely to become perfect at the task, rather than seeming human. But this can be circumvented easily with a GAN, (Generative Adversarial Network). Not only have the AI train to get better, but also have it train something that detects AI (discriminator, AKA detector). The detector will get better and better at detecting the bot, but the bot will get better and better at bypassing the detection. This means, that in order to detect the bot, Psyonix would need a detector better than the bot itself was trained on. This is virtually impossible. And if the detection singularity were to be reached, it would become literally impossible. *Anyone knowledgeable could train an AI like this with a bit of money to spend on cloud computing.* Point #3: "Preventing Cheats by Detecting False Inputs" - - - - - Some people also believe that making an anti-cheat, whether it be by detecting inputs/spoofing a game can detect these cheats. They cannot. The keyboard, controller, mouse, or whatever you use to send inputs legitimately, sends inputs in a very simple way. It is pitifully easy to send inputs virtually through the AI, in a way that perfectly recreates how these inputs are sent. It is as if a real controller were sending the inputs. Point #4: "Stop It The Same Way We've Stopped Cheating Before" - - - - - Some people believe the best way to tackle this is by detecting modified game files, etc. A pleasant reminder: The client-side is always insecure There is the server-side, and client-side. The reason Rocket League has been "unhackable" before, is that almost everything is dependent on the server, not the client. And the only data given and sent from the client, is necessary data of which not can be done with anyways. But AI only modifies the client-side, and the client-side is far more insecure. There is a reason cheaters always find a way to bypass anti-cheats in other games. There is a reason why piracy has always trumped anti-piracy. These are dependent on the client-side. The client-side requires positional, and velocity data, for the game of Rocket League to work. And these things are all that the AI requires to function. By sending inputs securely (which is not difficult at all), it is completely impossible to 'outdetect' cheats on the client-side. Not to mention, if the AI were to be trained using screen-data, rather than positional data, etc. No modifications on the client-side would ever be required. So no, having whitehat-hackers improve Psyonix's client-side anticheat would not fix this at all.

@@XerosOfficial what an effort you have to write this out man love it

@@XerosOfficial I completely agree with all of this. I never said anything about detecting an AI or trying to stop what is going on, just simply pointed out that this isn't the end and that these AI's will get better and better to the point where they're completely undetectable through adversarial training. There is no good solution to detect cheaters like that in the long run because client side anti-cheat is almost entirely pointless, and more likely to be thwarted by cheaters than modders, and it isn't doing anything that is physically impossible

Well, for now there's three weaknesses in Nexto that you can use against it: It has no aerial game, it has bad boost management, and its kickoffs are hardcoded.

@@theKiwii good you said "for now" because that comment wouldnt age well without it xD

Yes, it pops it up real high and consistently takes advantage of our momentum going in the wrong direction to take back possession of the ball.

Interesting, I wonder if that's because Nexto was trained against itself and it barely ever aerials, so it would make sense to prefer height

Because the bot has zero reaction time and can save everything. Power for human players is preferable because we have a reaction time. Well, in reality, top human players go after power and height. You not watch Mawkzy flicks?

Well i wouldn't risk it, he probably doesn't know how to recover from complex airplays

That’s a really good idea I’m shocked sunless didn’t think of. Just have them play 3 games without telling them what the deal is til the end of it

Agreed, Sunless took the approach of “can I play like a bot” instead of taking a pool of ~1700mmr players, randomizing them, and mixing the bot into those matches to see if ppl could spot the difference. But then again, this was for a YT vid so it’s not that serious lol

Yea he could do another test and not tell people. Giving that information tainted the results.

@@L3thalShad0ws7 Yeah, if it were serious it also wouldn't be vs a group of plats, it would be against the GCs getting "smurfed" on

No offense to Sunless, but they aren't anything spectacular for a gc1. From the opponents pov its harder to tell, but from his pov you can see he flipped the wrong way and got less power many times because of it, maybe even most of the time (which isn't that uncommon but he is just not as good as Nexto).

I don’t think you know what a good flick looks like man. Sunless has been playing this game for 9 years , since it came out in 2011 and he still can’t get to supersonic legand. Someone who can really do flicks is first killer. He’s so good compared to sunlesskhen

@@shadowyzephyr bruh this is the most unnecessary, self important BS comment. No one said he's a pro but can he not just be complemented without you coming in to add unneeded qualifiers to it? Shit comes off really insecure

@@dabloodynine exactly 😂

@@CBellRL CBell chatting shit?

Nexto usually respects rule 1 unless the ball gets too close to him

you cant say that 4 minutes after the video was posted. Hmmmm

It’s not as complicated as one might think. Human vs artificial inputs can be pretty easily discerned (at least with current ai’s used) based on how consistent they can hold the stick in some direction, how many inputs are done in a certain time frame, etc. I’d recommend looking into trackmania cheating (wirtual has some good vids) and how they discovered similar cheats in that game. Almost certainly would take a bit of time to implement, as I don’t think rl currently stores the inputs a player made during a game(very easy), and then they’d need to develop their artificial input detection system, which will mainly be a good amount of research, and some testing. Should be doable in 3ish months with a small team so they can still continue regular game development.

@@DaTimmeh yeah. I’ve seen some of those videos on the track mania cheating. I think that RL does store player inputs for goal replays, but there is no way that I, with a very basic understanding of coding, could prove or disprove that.

@@mik-ws3zw I think I've heard something like that before, could very well be the case.

@@DaTimmeh There are "some" inputs captured, but not enough to get any useful data on to train a network. A simpler method might be to just alter the game in a very small way that is indiscernible to humans, but would throw the bot for a loop. Something as mundane as adding a bunch of invisible actors with no collision physics either in or out of the arena, could/would cause catastrophic failure in the bots I/O.

@@KCM25NJL Ooooh that's a great one! If the bot is programmed to flick over opponents or do something to avoid them, adding invisible obstacles with no collision is genius. 👏🏻

It falls for the single jump if you're close enough, also you have to be very careful and very aggressive with your challenges

@@blobedm you clearly have no idea about how AI works

Then, make Nexto and the new AI battle it out and make Nexto learn from the AI, that would be _awesome_

there has been some tries of training bots by having them use the data from thousands of ssl replays. they didn't carry on as they are other better ways to have the bot develop airplay (like changing rewards)

Rolv did imitation learning from SSL replays, but it is not as good as Nexto.

@@swagie_ The point of imitation learning is it's good to start off with, then you train the bot against itself to improve it afterwards, which has been proven to work in other games.

If you watched the video you can see that Nexto does make mistakes

@@n.r.r1385 but it’s not that the bot doesn’t make mistakes, it’s in the fluidity of inputs, the perfection of steering when dribbling. Humans twitch ever so slightly with every movement, but Nexto’s maneuvering of the ball is always done with perfect mathematical fluidity, even if that fluid motion is a mistake. The inputs should be a dead giveaway and the server should be able to detect that with some clever programming.

There are plenty of solutions. The bots work in the same way that bakkesmod works which uses a form of DLL injection to rewrite the underlying libraries. Anti piracy tools exist to validate the checksum of libraries. Though, blocking bakkesmod would pretty much break the game for the comp scene

@@nyny Would be fine if you whitelisted BM. Frankly though, Windows should be signing all DLL's associated with an app install and preventing the hooking mechanism by requiring a valid signature for the hook.

@@KCM25NJL yes good reply. However bakkes is heavily driven by it's plugin architecture. That's why I didn't go into all of the details. They would need to make a submission and approval system and set up some kind of dev mode. All doable, but a bit of work

Cheating has been possible in RL in many different aspects for a long time, the knowledge has just stayed confined within specific communities, and now people are realizing. Cheating has never been impossible: there have been auto demo hacks, item hacks, and fake freestyling using replay editing or TAS, as well as a bunch of exploits which were reported first so Psyonix just gave the player a white hat and people didn't realize they were ever possible.

I've looked into the bots a bit, the 'impressive' part would have been reverse engineering nextos actions into controller inputs. Controller inputs aren't sent to the server, only some data like pitch / speed / etc. However, the client app can detect if there are controller inputs, which the bot creators would have to implement for any hope at going undetected, because the client can communicate with Epic still. There are solutions for Epic, it would also block bakkesmod though which is pretty much game breaking at comp levels.