thanks for mentioning closet hackers, esp togglers, etc. this is an epidemic in high elo premier right now. its so obvious but reporting them does nothing
whats funny is that I met 2 players in the past 2 weeks with cs2 bans on mass report bot (idk how it works really or even if the source is accurate but it is none the less frustrating).
I dont know if you know what "closet" cheating is, if you're also going to say it's "obvious". If it's obvious, that's not closet cheating. That's semi-rage at best.
EDIT: Others have pointed out it could mean 99/100 cheaters are banned, not that it has inaccuracies in detection. If that's the case then no problem. I'm less than 30 seconds in but want to stress that banning at 99% accuracy means 1/100 bans are innocent players. So if you ban say 100,000 cheaters you're banning 1000 innocent players too. That's actually not good enough.
A valve employee was also banned in that ban wave. thats how the reverted so quick, normally takes valve about a week to rollback false bans because they need to confirm it is false or not. But a valve employee getting banned is proof enough that it was false ban wave.
Hi, as an AI researcher I must clarrify some innacurracies about AI models. I will do it chronologically, as they appear in the video. 8:20 : " Nobody [...] has developed an AI that can think independently in any meaningful way " this is absolutely true. But it's important to understand why. The keyword here is "think" because an AI does not think. It recognizes patterns and imitates data. When ChatGPT generates an answer it will give what looks like a good answer. Since it has seen way more text than any human can read in a lifetime, the answer will seem well thought out. In reality, ChatGPT will take the question, generate the next most likely word, and to that till it reaches the end of what looks like a good answer. That's why, when asked to use logic (math for example) an AI will fail. When given an equation ChatGPT will give what looks like a good answer and will see very little difference between an 11 or a 12. 8:40 : You explained Hallucination well. However, this is not comparable to the task of cheat detection. A model is trained on a specific task, and there are many different tasks. Cheat detection is a classification task. Meaning that the output will be either 0 (not cheating) or 1 (cheating). Hallucination is a specific type of error that happens with natural language generative models. When it's used outside of that scope it's just an error mislabeled as hallucination. In the rapidly evolving world of AI, these terms missuses are very common. 9:30 : Machine Learning. Here is a big shortcut, not sure if it's an error or to simplify the video, but if you're interested here is why it's not completely true. There are different levels of AI. First, there is AI. That's anything artificial that simulates intelligence. For example, NPC AI in video games, where the algorithm is coded by developers. Secondly, inside AI there is Machine Learning. Here the algorithm is coded by someone but the values are found by the computer. For example, you have a flower and want to know the species. With a machine learning model, you will have to measure yourself each part of the flower and give yourself information about the color for the model to answer. For machine learning the data needs to be of a specific structure. This is advanced statistics. Lastly, inside Machine Learning there is Deep Learning, where the structure of the algorithm is made by someone, but how the decisions are made is found by the model during training. There is a lot more freedom in the type of data that can be used. For example, pictures. In the flower example, a simple picture would be sufficient to determine its species. ChatGPT uses Deep Learning. 9:40 : "We can never feed it perfect data" Yes, but does it mater? You're right about everything you said after that. However, in AI there is a technique called "Unsupervised Learning" that learns from data without human supervision. The model will search for similarities in the data and form groups. Players' behaviors can be extracted by the model to form said groups. Then a human can identify what the groups correspond to. If a group is made of cheating players, then when the model categorises a player in that group that means he's cheating. Obviously, I have heavily simplified the process, and it's, in reality, way more complex, but the main idea is there. Anyway, great job and great video ! Loved your insight on the subject.
@@lawsen3719 What a sad response to a genuinely interesting and structured comment with actual insight into the processes at hand. Took me a minute or two to actually learn things about AI I did not before. So thank you OP, for taking the time to write this so us plebians could learn something.
The issue with AI detecting cheaters is that the server data cannot be trusted unless the client is fully secured. A lot of cheats depends on looking up what is in the memory and showing that data in an overlay (I think)
Honestly the best way of going about it is through "paying" via in-game rewards kind of like CS's Overwatch had. Or even incorporating it as some sort of intern role. I honestly do not expect companies to hire people to do this sort of thing, its an immense expense to have enough manpower for it.
@@techelitesareadisease8816still, csgo makes enough scrooge mcduck money from cs to hire atleast 100 people to do this. It would not fix the issue by any means but atleast people like the dude from the start who was botting overwatch bans get punished. At what point do you just say fuck it and hire these people. Short summary of my opinion: Valve needs to hire people to ban cheaters - overwatch is useless because it can be abused by cheaters - Valve has no excuse of expense considering how much csgo makes them every month - Big cheaters with 10k inventories need to be banned but valve isnt doing shit.
Remember when you could get banned from og mw2 / 3 for someone accidentally pressing the "max prestige" button in private zombie / modded lobbies. Remember when you didn't get banned from black ops 1 for max leveling using theater mode then blowing yourself up with a grenade in a normal lobby... Now we can get banned if an AI decides we're too good not actually using cheats.
It makes sense. In CSGO bots would spin around in one spot constantly. AI gets trained that kind of behaviour is of a bot. Now if you try and copy previous bot behaviour by spinning fast AI anti cheat will ban you. Ai is only doing what it was taught. Edit: Forgot to mention it is not the high dpi that's the problem. Spinning around constantly is.
But this banning is ridiculous because people aren't using any cheating software at all, they're just moving their mouse around quickly. Off topic, if you attack a teammate while moving your mouse quickly, you will be banned almost 100% of the time. This is the downside of AI anti-cheat.
The video shows you have to do it a lot before you actually get banned... Always be sceptical when people claim false bans.. Cheaters always say they were falsely banned. There was a "false" ban wave in Hunt Showdown recently which I don't think was a false ban wave. The people banned all claim this Hunt Stats app was the cause but why would anyone download some crap stat tracker that tracks basic info like kills and extracts? Who heard about?.. The rumours are this app was pulling chat logs in game and were being used to ping a players location or something to that effect. None of this was disclosed by the app maker themselves.. There was nothing officially claimed it could do anything to get you banned. I'm of the opinion that either the app was a trojan horse to catch cheaters or it was made with an embedded cheat and the only people who knew about it were cheaters. Who downloaded it used it and got banned. The day the ban happened everybody said games were so much fairer. As soon as the ban was lifted it's back to cheaters in every single game.... So it wasn't really a false ban.. All these people who are going into CS 2 and spinning their mouse around to get banned are probably cheaters trying to create situation to make it look like false bans....@@perfsakuya
@@SynWOSall criminals breathe and drink water. TO AVOID BEING A CRIMINAL, DONT BREATHE OR DRINK WATER. Yes, spinning is clearly spinbot weighted, but cheating bans is not something you can "mistake" with.
9:45 The trick is for valve to download cheats themselves and create their own sample data from it so they can be certain that the data they are training on is actually a cheater.
I left the AI field few months ago. I got sick of upper management changing directions because the saw or heard someone else is doing something new or using something new. I swear in 8 months I wrote 6 lines and probably went to 150+ meetings. AI is a dead money pit. 99% of these projects are scams
A never ending battle. AI will make it interesting. Cheat devs will have to get crafty. They make huge money and give themselves backdoor access to people's stuff. They won't give up easily.
Unless cheating in games can bring repercussions in real life like in South Korea, the only thing anti cheat can do is to make cheating so tedious so that the process to get cheat to work is enough to discourage most of cheaters. But this also means there will still be cheaters who spend over 100 dollars per month just to cheat.
This is why I'll never use voice chat in GTA again. They just started to use AI to scan for 'toxic' behavior. I can only imagine how many false bans are going to be raining from heaven now
This was such a predictable problem. I love how in the previous interview they say 99% and are not worried about false bans, or fingerprinting, but how am I supposed to trust that if I do a flick or micro adjustment I won’t get banned. I have to worry about getting a false ban and cheaters get free rein. It’s unfortunate and I hope something is done.
In a way I can respect that they are trying to do something unique with the anti cheat and with the sub tick etc etc, but it’d be so much better for the health of the game and so much easier if they’d just get a damn kernel level anti cheat and 128 tick servers. Like 99% of cheaters would be done overnight if they got a good kernel level anti cheat like riot. I’d trust valve with my stuff any day way more than riot
I think what people are forgetting is the fact that training AI cheats (the counter to AI anti-cheats) in a race against one another will be both difficult and time consuming. There may be a large market for cheats which invites some "experts" to place their claim in the market and develope AI, but generally, the experts will be working at Valve. Those people will have unmitigated access to any data or computational requirements necessary to fight this never ending war. I think the back and forth with be reduced from weeks at a time to hours. This will mean that people will have to buy new cheats every few hours. They will lose trust in the cheating system, or they will lose too much money and be unable to keep bleeding their wallet. If attrition is what cheaters want, they will be sorely mistaken.
I got banned after my g pro sensor went crazy and spun tf out of me. I was getting a lot of "play stupid games, win stupid prizes" as if I was supposed to factor in a possibility of my hardware getting me vac banned.
Addressing cheating should be the #1 priority of publishers and developers. The gaming industry should have its own version of the Manhattan project to address this problem.
You want an absolutely perfect system for detecting cheating/rule breaking in games? Go look at iracing, community reporting, but EVERY report has to be looked at by a person. It takes longer, but Iracing support is actually the best support I’ve ever dealt with. Truly setting the bar in all cases.
Man, i don't wanna seem like an asshole here.. but you could really use a pop-filter on your Boom-Mic, y'know? It'd completely remove all the tongue-click sounds, and would go a long way towards cutting down on the way you sound like you're gasping for air like a man on the verge of drowning between every sentence. Despite the fact that I am *100% sure* that were i in the room with you, i wouldn't be able to notice even the slightest hint of either of these two things. Also, just wanna say; Once again, an _absolute_ banger of a video, man. I fucking love your new post-Tarkov content. Honestly, in my opinion, you've yet to miss the mark even _once_ with these more 'investigative journalism' style videos. Cheers, g0at. Glad you're sticking with it and making great pieces.
bullsh*t ! The Ai anticheat just works fine. Of course all cheaters who got banned would come and say " buuuhuu i just play with 50.000 DPI ! im not cheating :("
As a software developer and AI researcher, i was heavily critical of AnyBrain pointing out most of the same, valid points that you made in this video. Bravo; great video and very detailed. Keep up the content!
Smurfing is cheating... Just like it' cheating for a heavy weight to fight a bantam weigh or for a Man to Pretend to be a woman and kick the shit out of women in female UFC.
Some things that you said are a bit off base: 1) Purely client-side cheats like god-mode or esp are not detectible by anti-cheat, how would they be deteced? - This is mostly not a problem for server authoritative games. There is no god mode that a client can run if a server is the authority on who takes damage. If client 1 shoots client 2, client 2 can deny it all they want, but the server doesn't care. The server is running a copy of the game and their copy is the only one that matters - ESP is possible when a client is an observer of an object. A client can only know something exists if they are told it exists. A client can only know where an object is if they are told where it is. Only observers are told this information and secure games manage observers dynamically and with server authority. ESP is only possible when a client observes all objects or when it is possible to observe all objects by force - Basically, this is a game security problem, not an anti-cheat problem. The solution to games that have security problems is obviously just to learn the behaviors that players with these cheats show. As you said, if you can't get this data then it's impossible 2) The fingerprint isn't possible since the inputs can be masked - Client side inputs don't matter if they aren't sent to the server. The finger printing should work just fine on the information that is sent to the server since most cheats aren't changing every input that is being sent to the server - When cheats decide to mask those inputs being sent to the server, they would need to change all of their inputs. There would be no original walking or aiming input that is sent to the server at any time in this case. The user of the cheat would be able to avoid fingerprinting, but would feel that the game is less responsive since their exact inputs are being manipulated in some way. This could involve value tweaking, input delay, or false inputs - Fingerprinting obvious can't track people that mask all of their inputs, but it's at least another line of defense since many cheat users would probably not enjoy the masking
You have probably already considered this but I feel like taking your channel in the direction of focusing on anti-cheat in video games and doing case studies on successes/failures throughout history would be really insightful and interesting.
valve is the one company that could afford to buy a sky scraper and have it be anti cheat tower and just have 15000 employees review cases or watching suspected people in games for toggling
Not seeing the client side part.. Going way back, cod 1, cod 1 uo, we had this screenshot mechanic literally taking screenshots of the screen of the player and we banned loads of people based on just that. The same bots that did not get detected by punkbuster at the time got rekt by just reviewing the screenshots saved in a folder on your server. I understand privacy and what not. But "rootkit/malware" kind of anticheat should be embraced if we want to get rid of these people. Hell back in the days there were servers specifically for people who wanted to cheat. Maybe i'm just getting old. Maybe i'm over simplifying things i don't know. But heck old internet was better.
As an ex cheater on source, it's quite difficult to be detected by you teammates or the other team. None the less at the end it's only a game, a closet cheater who only use Wallhack can be defeat by any player. And don't forget them the industry including valve CREATE this issue. Before everyone was playing on private servers admins and moderators have to mange cheaters. A server well maintain whit good staff was I sign of quality, you have to respect your teammates and play accordingly to server rules.
They don't need to go through every single high DPI ban to determine if the ban was correct, all they need to do is have a system in place that checks if the player got a kill while spinning
@@garakthetailor pirates an online game? Huh? Also no, not only poor people cheat, you can easily just get a new account and boot the game back up if you get banned on a free game. YOUR ECONOMIC CLASS HAS NOTHING TO DO WITH IF YOU DECIDE TO CHEAT IN A GAME
@@garakthetailor yeah, if a piece of software costs money, people are going to pirate it. Making an ONLINE ONLY game paid instead of free would not increase piracy levels astronomically like a singleplayer title. Also, if a game is free, people are not going to pirate it at all, no shit piracy levels would go up, but in a game like CS, the profit loss from piracy wouldn't matter because they're already shipping the game for free to begin with? Meaning they don't profit directly off installs already?
The issue is match making servers. Back in the day, we had dedicated servers only, like on Modern Warfare with communities hosting their own servers. It was easy for players to police their own servers. When they introduced match making in MW2 and didn't allow for dedicated servers the cheaters took over within a week. Match making took over for game releases around that time and the cheating issue exploded to what it is now.
It’s almost like implementing a kernel level anti cheat into a competitive fps known for its cheating issue would not only sort out false positives but detect game hacks with higher efficiency than any other method and would be hugely positive for the community. I give valve my buyer’s info, I’m not worried about them, as a private company, fucking me over via kernel access.
you shouldn't be worried about valve fucking you over. with a kernel level anticheat you essentially create another door to your system. how secure that door is is up to valve. besides that they won't go kernel level. that would go against all the things they are planning with linux.
stuff like this is the reason why I dont play online competetive shooters anymore. my gaming experience has been amazing and much better now. the whole scene is just.. kinda a dumpster fire now anyway.
It can be frustrating how much speculation can be involved in this field, but even educated guesses can be helpful and interesting when dealing with these black boxes, so thanks!
My very recent game which was last night we had a spinbotter with an AWP on the other team just killing everyone in just 15 seconds of every round (also has rapid fire so the AWP is just like an auto-sniper). Why is VAC live not able to detect that??? :(
Closet hackers is my problem, I don’t even run into the obvious troll hacker often at all, it’s the closet cheaters, those kids with the low self esteem or w.e reason why they end up cheating. I’ve let go of competitive games. Especially games that can run on a potato like cs2
bro, am I the only one who realizes that 99% is actually hella low? in a playerbase like CS, it would mean that 7900 ppl would be banned for nothing. wtf
ESP has very obvious patterns that an AI could detect, like aiming straight at another player through a wall or the percentage of successful pre aim (aiming where a player will appear in the future)
For the cheater to do any multiplayer action in a multiplayer game they have to send some input to the server so that the server can update the game state for all the other players. That is the input you analyze, the cheater can't fake this data because then they wouldn't be able to achieve their goals ingame
You're doing gods work for the competitive / pvp gaming industry. I've seen too many people stating that Anybrain would fix everything, but when I went to observe their website a year ago, and looked into the available information about them I came to the same conclusions as you have. You've of course gone a mile beyond in showcasing the bullshit they are peddling! All the glory to you, and I can only continue to follow you and see how far you can go!
The only way to feed the ai correct data is if you pay people to play on your hardware, with and without cheats and make sure they can't do anything other than play the game the way you intended them to play. You record that gameplay and feed it to the ai. But you'd need thousands of players at different skill levels (with and without cheating) to get an accurate portrayal of the player base. You could also choose to not ban anyone so that cheat software doesn't evolve to evade your anti cheat. Then you mark everyone that for sure is cheating with your conventional anti cheat. Try to find each and every cheat on the market and play with it to see if the cheat gets marked. Then train the ai on the gameplay data from all the cheaters and non cheaters. And you'll have a fully correct trained ai. Assuming people didn't get falsely marked.
Every human has blind spots, while this is only one way to "Fingerprint" this would be quite obvious, as a player detecting in specific area's of their known screen area, will show a perfect heatmap. I am sure there is far more to this, such as when using Aim trainers, I suck at the upper half of the screen, while I a apparently pretty good at the bottom half, etc. Hopefully this isn't just smoke being wafted gently up our behinds, Which I doubt as AI is at an extremely high level already, so it is likely they know all this basic stuff mentioned.
In regards to AI requiring humans to label examples of cheating and legit play, that is not entirely true. AI can work in an information vacuum and kinda brute force its way through the problem. The more you help it by giving it corrections the less brute forcing you have to do. The various forms of cheating ARE objectively distinguishable from legit gameplay, almost by pure definition. In other words, cheating either directly modifies the game in some way which can be detected, or in the case of cheats that "cant" be detected they modify human behavior such as having a wallhack would change how you play and behave. An AI would in theory be able to separate out these various groups without actually know which group is which and without having a single example of what a cheater is by purely being feed the replays of every single game. At this point you would have what is basically a sorting machine that groups "like with like" at which point you can have a human come in and analyze which of these groups represents cheaters and even what kind of cheat they use. Of course, doing it this way would work but also be one of the slowest and dumbest ways to accomplish a similar thing but its more intended to demonstrate that it is possible.
Was banned from CS 1.6 two decades ago. Still dont know why and when asking they only told me that they dont know/cant tell me why and that they cant do anything. If i was cheating, i wouldnt make any sound, but till this day i still dont know why. Was it because i changed the config and had some "quick buy" options (pressing any num key to buy whole kits of gear), or because i imported some spray logos...IDK🤷♂
This is nothing new for Valve or CS. I got a false ban 7 years ago in CS:GO, playing only casual matches and zombie escape servers, despite never cheating in a game in my life. There's been nothing I can do about it since.
That company provided you a research paper on how their tech works, but rather than try to understand it, you just seemed very condescending for them having even tried, but it just not being enough in your mind? And why should they go more into detail for you? So you can make a quick buck and get some publicity? Pretty petty reaction to not getting a phone call. They're a tiny startup trying to do something new in the space, but it's apparently not good enough for the entitled UA-camr. Also smurfing should be bannable, and in that case, what's the issue with the ban at the start of the video? While some bans might not be justified, being strict on suspiciously good fresh accounts like that is an extremely effective way of preventing cheaters. Those bans may only be temporary too until someone manually reviews it. But yea smurfing just creates more work for people manually reviewing reports, so it's perfectly reasonable to be banned for this. Also at the end when you talk about them needing to validate the clients input came from their hardware directly, that's not possible. You receive data from the client and validate it from the server, no software on the client will ensure it hasn't been tampered with. Never trust the client, it's just a rule. Solutions like TPM will help prevent tampering though, but most games still don't enforce this as strict as they can because it would limit their player base too much.
You know when you feed AI an image and it will throw you what it thinks it is and the percent confidence it has in that prediction? Say Duck (99.7%)? The fact that the AI can't be completely confident, as well as the fact we can see it being completely wrong sometimes, should basically kill the idea of relying on AI as the arbiter of cheating here and now. AI will get better and is a good way to expand the net and find suspicious people, but we need humans reviewers to at least look at appeals and hand out final decisions.
People will say that cheating is an issue, and that the solution is for the devs to make infrastructure to stop it, but time and time again, we see that devs and players both working to stop it is the best solution. CSGO’s Overwatch was legitimately useful, the issue was that cheaters figured out how to game the system. Better bot protection and a more open platform for the community to examine the data themselves would prevent obvious flaws that existed previously from going unnoticed, hell, they could go hand in hand. I think the use of AI is definitely a good idea, in so much as catching obvious cheaters quickly, but they do need to introduce some human element into it, especially considering it has the ability to VAC ban. Perhaps even an Overwatch LIVE system could even be implemented, to give CS2 streamers and players something else to do. Obviously it can’t be 100% real time, but spectators of competitive already have about a 1 round delay anyways, so they could easily just import that. While I think Valve should definitely be utilizing AI to empower themselves, they should also be utilizing it to power the community as they did in the past. It’s also strange seeing Valve move from VAC bans having a near 0 false positive rate to a large number of them being false positives as of recent. Definitely represents a mood and/or ideological change in the development team imo.
overwatch had it's issues. from 2015 to 2018 you needed 11 unique reports to have your recent match(es) be sent to overwatch. "a game a day keeps overwatch away" was the motto. from 2018 onwards they implemented some sort of scoring system to automatically send matches to overwatch. something a long the lines of -> a few games with ~100 score and you're in overwatch. in 2020 i think? they started to automatically send rage-aimbotters into overwatch. this is where overwatch lost its purpose. it was filled to the brim with cases that could have been handled by software therefore making room for cases that require actual human intervention aka the closet-cheaters. i agree that overwatch isn't a bad idea. but the conditions to get sent into overwatch needs some refining and it's near useless when all you get is ragecheater after ragecheater.
7:20 I wonder why with all the layoffs and them not hiring 8:17 pretty sure if we used 150th of the military budget. Lots and lots of cheaters would be gone
Nice video, thanks for the content. I just want to point out a mistake you make on my point of view: you state that training AI with already identified cheaters will only allows to detect the same kind of cheaters. BUT: you forgot about the fact that if a human use let's say 4 variables to identify a cheater, when the game registers 12 variables. If you train the AI giving it the whole dataset (so 12 variables), then afterwords it will be able to identify cheaters based on the others variables that the human can't see/handle. A good comparison would be AI trained to detect diseases with cameras. We use hyperspectral camera that catches more or less the whole light spectrum. Then a human identifies the disease on each picture, but the human only uses its eyes, so only a portion of the light spectrum. But the AI now know that the other spectrums of light are like this or this when there's a disease. So in the end, AI will help to catch knew type of cheaters, if developped properly :)
So strange to attempt to solve this incentive problem without modifying incentives. The gaming industries should be campaigning to make being a cheater in video games be a real world crime with real world consequences: fines, community service, jail time; this also benefits from the requirement of enforcement being as delicate as it is in real life, beyond any reasonable doubt with unions and an appeal process.
This is an interesting point, and I wonder if it likely already is a crime since you're using a company's network and infrastructure in a way that bypasses their security.
Cheating is Illegal in China... The country with the most cheaters... You can't rehabilitate cheaters... Once a cheater always a cheater and cheaters reproduce with people creating more cheaters and cheaters reproducing together creating even stronger cheaters... It's a virus...
While I'm absolutely for punishing cheaters with fines, jail time and so on, this idea is very very sippery slope. While it should be illegal to tamper with software to gain an advantage over another person in an online game, I'm certain that the fucked up gaming industry giants would also try to outlaw modding as well. We've seen companies try this shit in the past (rockstar for example). This would bascially kill any unoffical modding which is a huge loss for the industry.
Secure clients are not really ever going to be an actual solution unless you have to go to a physical location with security and human surveillance, show ID, and play a game on the company’s computer like how we do professional testing. Sever-side validation will remain the primary mechanism to validate user input. Checking that humans are pressing physical keys isn’t really the point anyway, for example disabled gamers may need virtual inputs or whatever. Cheating is a social problem and thus does not have a technological solution imo. Humans have cheated long before games were on computers and we will continue to do so. I think chains of trust (I.e. you vouch for me and I vouch for my friend) are a promising way to find groups of players that you want to play with and it can be automated and scaled with more technology. That may not work well for competitive games though. But to be honest we can’t even stop people cheating in physical sports either so 🤷♀️
While you provide interesting feedback and views on the topic and this product, however I believe you're too much stressing about the client side here. While I agree that the communication from Anybrain could be better and a little more transparent, I think you're simplifying this topic a bit too aggressively. Additionally, less is literally more for AC developers. For example, this kind of AI-based detection is solely based on human-factor. It doesn't need to know how ESP looks like for the user or what inputs the user gives, but it needs to distinguish the behavioral patterns between different cheat features and legimitate users. And as you said, this requires a great amount of training material that has to solely consist of already known cheaters (& cheat features) and non-cheaters. However! This doesn't make it impossible, but it does mean that an AI-based anti-cheat detection system WILL HAVE TO work tightly together with traditional anti-cheat detection tools to get the required training material. Over time any AC client will be able to gather the required data, as long as the AC developers are conscious of the technical specification requirements of training material. There are already a lot of studies available from other fields that can be utilized here. I'm sure this will be only a matter of time until we've a working solution available. Perhaps Anybrain will fail, or perhaps not, but there will be something similar available in near-future.
Bring in pros to the building, closed hardware, use them as a baseline, and ban the above. If a false ban is reported, they must prove it, and if able they become part of or are the new baseline. Same for hacks, use actual hacks on closed hardware so the system knows what its looking for when they are used. Human oversight will always be needed in the next decade.
I've played with Circuit players, they can easily play like regular people when they aren't playing for positions. Your argument relies on professionals being incapable of playing like players, which is not only insulting but most importantly ludicrous. @guymontag5084
If real, I would assume their looking at the data sent from the client. While the data could be doctored, it would probably show signs of tampering. (Much like photoshopping a photo.) They could be looking at other factors, like latency between sent vs resaved data. Variations in the data. Average steps in the data. While it may seem like an impossible task. This is the same issue people face with virus, malwares, extra.
I am going to get pitchforked here, but I am going to write this wall of text anyway, maybe someone can see through the blinders because of it. If one person would, that makes it worth my time. I am pretty sure that pro players in CS GO were cheating on all LAN events in past years. Valve did absolutely nothing to stop it, because they do not care about integrity of game as long as it is making them obscene amount of money. They let kids gamble, let people ruin their lives for pixels. Goat already touched this subject in 2 part video series, BRAVO! Nothing else matters for them. I believe Valve FINALLY found solution to cheating problem in Counter Strike. The game is the CHEATER now and it doesnt matter that Billy has walls or aimbot in grand schema of thing because it just doesnt matter anything in result of match. Bare with me here: 1. AI powered driven big data matchmaking 2. AI analyzed chat and voice chat in games to see how players react in match by game cheating on them or they accuse their opponents of cheating - likeliness of player quitting game session 3. AI controls interpolation (peeker's advantage), crosshair movement (micro adjusts your shots to hit or miss opponent), change recoil patters, hearing range for footsteps, flash bang effect duration,... I have seen all in my games and there are quite a bit more in there. 4. every player in game has good and bad moments within each match or at least in series of matches to make him "feel" good at the game to continue playing 5. cheaters are irrelevant because game can "make" them die by legit player just by manipulating game itself = cheating problem solved? 6. players are unaware of their own true skill and basically spoon fed dopamine and rank without progressing in game as they should - if you play CS for 18+ you just know Feel free to comment about your game experiences. How do you like new CS2 compared to CS GO, CSS, CS 1.6 or even older version. I started with CS 1.0 and I am pretty sure this one is by far the worst CS ever release.
as someone that's been making hacks since like 2000 I laughed so hard at that 99% naw man you can 100% keep hackers out of games now they just don't because they need someone to sell the game to after it's past it's prime. Same with botters. Look at blizz they print money every 6 months. Why 6 months? Because it gives botters time to buy back in and work for 6 months. Hell the turn around rate on blizz bots is 100% because it's to good of money for both sides to pass up. Just think you can sell a game to a hacker at lest 4-6+ times with a botter they will always turn around and buy back in. It will never stop you'd be dumb as hell to 100% stop hackers and botters in your game.
I will say, g0at, that synthetic data is the long term solution to high quality data. Maybe we don’t know what a cheater is from Overwatch, but, if we can successfully synthesize counterstrike games of sufficient quality (or images of ducks, or well-argued essays, whatever), then we can significantly increase the accuracy of these systems to distinguish between players and machines. But, yeah, obviously if anticheat datasets can synthesize realistic human gameplay with machines, so can users.
If I had learned anything from my machine learning class in college, machine learning can find things humans can't but only if the algorithm and the training set is suitable. A poisoned dataset is normal for these kinds of problems for the AI, and that the model itself should be able to figure out. If it doesn't work, its either a bad algorithm/implementation , dataset, or training power.
it would still be nice to have a basic anticheat that will ban obvious cheaters quickly and effectively, not requiring a manual review or player admin etc
AI: “My job is to make sure NOBODY cheats”
*Bans everyone*
Haha, awesome
No one can cheat if they can’t play the game
Judgement day
Ok cheater
😂😂😂 THE LARGEST ISSUE IN CHEATING IS PLAYERS THEREFORE WITHOUT PLAYERS THERE NO CHEATERS! it’s going full skynet!
bro became an authority on cheating in gaming
all hail the anti-cheat supreme leader
For real. Dude is finding cracks in the matrix.
She's really the furthest thing from it. The only thing wrong with anti cheats is that humans are involved that makes them flawed
@@RekySai You're a furry. Nobody asked.
The wiggle. 😆 🤣
thanks for mentioning closet hackers, esp togglers, etc. this is an epidemic in high elo premier right now. its so obvious but reporting them does nothing
lots of them aren't even closeted lol. 20k elo premier both sides openly cheating and talking about it
its been an epidemic all over
Honestly, like wow you're suddenly nailing headshots and showing game knowledge that you didn't have?
whats funny is that I met 2 players in the past 2 weeks with cs2 bans on mass report bot (idk how it works really or even if the source is accurate but it is none the less frustrating).
I dont know if you know what "closet" cheating is, if you're also going to say it's "obvious".
If it's obvious, that's not closet cheating. That's semi-rage at best.
EDIT: Others have pointed out it could mean 99/100 cheaters are banned, not that it has inaccuracies in detection. If that's the case then no problem.
I'm less than 30 seconds in but want to stress that banning at 99% accuracy means 1/100 bans are innocent players. So if you ban say 100,000 cheaters you're banning 1000 innocent players too. That's actually not good enough.
especially in a game where many of the players have thousands of dollars in skins that get tradelocked. it’s a lot more than an account loss.
You do realise though, a 100% success rate will always be impossible when you are unable to have all data available.
your right, you aim for 99.99999999999%. 99% is nowhere near enough but, 99.99999999999 is probably overkill.@@BAGG8BAGG
@@BAGG8BAGGwhich is why this will never work
@@BAGG8BAGG I'm not sure what your point is as you've just stated an obvious fact. Do you think falsely banning 1 in 100 players is acceptable?
A valve employee was also banned in that ban wave. thats how the reverted so quick, normally takes valve about a week to rollback false bans because they need to confirm it is false or not. But a valve employee getting banned is proof enough that it was false ban wave.
i mean he definetly could be although it would be pretty reckless lol @@forbidden-cyrillic-handle
It was the unlucky one that downloaded the cheats, so the team could look at them, to develop the anti cheat against that cheat
xD
Hi, as an AI researcher I must clarrify some innacurracies about AI models. I will do it chronologically, as they appear in the video.
8:20 : " Nobody [...] has developed an AI that can think independently in any meaningful way " this is absolutely true. But it's important to understand why. The keyword here is "think" because an AI does not think. It recognizes patterns and imitates data. When ChatGPT generates an answer it will give what looks like a good answer. Since it has seen way more text than any human can read in a lifetime, the answer will seem well thought out. In reality, ChatGPT will take the question, generate the next most likely word, and to that till it reaches the end of what looks like a good answer. That's why, when asked to use logic (math for example) an AI will fail. When given an equation ChatGPT will give what looks like a good answer and will see very little difference between an 11 or a 12.
8:40 : You explained Hallucination well. However, this is not comparable to the task of cheat detection. A model is trained on a specific task, and there are many different tasks. Cheat detection is a classification task. Meaning that the output will be either 0 (not cheating) or 1 (cheating).
Hallucination is a specific type of error that happens with natural language generative models. When it's used outside of that scope it's just an error mislabeled as hallucination. In the rapidly evolving world of AI, these terms missuses are very common.
9:30 : Machine Learning. Here is a big shortcut, not sure if it's an error or to simplify the video, but if you're interested here is why it's not completely true. There are different levels of AI. First, there is AI. That's anything artificial that simulates intelligence. For example, NPC AI in video games, where the algorithm is coded by developers.
Secondly, inside AI there is Machine Learning. Here the algorithm is coded by someone but the values are found by the computer. For example, you have a flower and want to know the species. With a machine learning model, you will have to measure yourself each part of the flower and give yourself information about the color for the model to answer.
For machine learning the data needs to be of a specific structure. This is advanced statistics.
Lastly, inside Machine Learning there is Deep Learning, where the structure of the algorithm is made by someone, but how the decisions are made is found by the model during training. There is a lot more freedom in the type of data that can be used. For example, pictures. In the flower example, a simple picture would be sufficient to determine its species. ChatGPT uses Deep Learning.
9:40 : "We can never feed it perfect data" Yes, but does it mater? You're right about everything you said after that. However, in AI there is a technique called "Unsupervised Learning" that learns from data without human supervision. The model will search for similarities in the data and form groups. Players' behaviors can be extracted by the model to form said groups. Then a human can identify what the groups correspond to. If a group is made of cheating players, then when the model categorises a player in that group that means he's cheating.
Obviously, I have heavily simplified the process, and it's, in reality, way more complex, but the main idea is there.
Anyway, great job and great video ! Loved your insight on the subject.
I aint reading allat
@@lawsen3719 What a sad response to a genuinely interesting and structured comment with actual insight into the processes at hand. Took me a minute or two to actually learn things about AI I did not before.
So thank you OP, for taking the time to write this so us plebians could learn something.
@@lawsen3719Then stay dumb. 😊
Super interesting comment thanks
The issue with AI detecting cheaters is that the server data cannot be trusted unless the client is fully secured. A lot of cheats depends on looking up what is in the memory and showing that data in an overlay (I think)
I'm glad there is someone who is being critical and realistic in regards to anybrain's marketing and not just eating up whatever they say.
why are you so glad about it? scared to get banned? ;)
I was on the counterstrike reddit and was watching this happen live, just crazy
Just hire that one jobless friend who keeps on saying "broo if these companies would hire me to ban cheaters all day, there would be no more left" 😂
That would be a dope job especially if they played the game
Honestly the best way of going about it is through "paying" via in-game rewards kind of like CS's Overwatch had. Or even incorporating it as some sort of intern role. I honestly do not expect companies to hire people to do this sort of thing, its an immense expense to have enough manpower for it.
@@techelitesareadisease8816still, csgo makes enough scrooge mcduck money from cs to hire atleast 100 people to do this. It would not fix the issue by any means but atleast people like the dude from the start who was botting overwatch bans get punished. At what point do you just say fuck it and hire these people. Short summary of my opinion: Valve needs to hire people to ban cheaters - overwatch is useless because it can be abused by cheaters - Valve has no excuse of expense considering how much csgo makes them every month - Big cheaters with 10k inventories need to be banned but valve isnt doing shit.
I'll do it, Valve hire me
With how much my friends call BS on legit kills, there would be a lot of sad gamers.
Remember when you could get banned from og mw2 / 3 for someone accidentally pressing the "max prestige" button in private zombie / modded lobbies. Remember when you didn't get banned from black ops 1 for max leveling using theater mode then blowing yourself up with a grenade in a normal lobby... Now we can get banned if an AI decides we're too good not actually using cheats.
Better then what we used to have
It makes sense. In CSGO bots would spin around in one spot constantly. AI gets trained that kind of behaviour is of a bot. Now if you try and copy previous bot behaviour by spinning fast AI anti cheat will ban you. Ai is only doing what it was taught.
Edit: Forgot to mention it is not the high dpi that's the problem. Spinning around constantly is.
But this banning is ridiculous because people aren't using any cheating software at all, they're just moving their mouse around quickly. Off topic, if you attack a teammate while moving your mouse quickly, you will be banned almost 100% of the time. This is the downside of AI anti-cheat.
The video shows you have to do it a lot before you actually get banned... Always be sceptical when people claim false bans.. Cheaters always say they were falsely banned. There was a "false" ban wave in Hunt Showdown recently which I don't think was a false ban wave. The people banned all claim this Hunt Stats app was the cause but why would anyone download some crap stat tracker that tracks basic info like kills and extracts? Who heard about?.. The rumours are this app was pulling chat logs in game and were being used to ping a players location or something to that effect. None of this was disclosed by the app maker themselves.. There was nothing officially claimed it could do anything to get you banned. I'm of the opinion that either the app was a trojan horse to catch cheaters or it was made with an embedded cheat and the only people who knew about it were cheaters. Who downloaded it used it and got banned. The day the ban happened everybody said games were so much fairer. As soon as the ban was lifted it's back to cheaters in every single game.... So it wasn't really a false ban..
All these people who are going into CS 2 and spinning their mouse around to get banned are probably cheaters trying to create situation to make it look like false bans....@@perfsakuya
@@perfsakuya Again do not spin around constantly mimicking the behaviour of CSGO bots.
@@jubbin4849 I agree. I am just saying now we know what the problem is don't do it until we know it's been fixed.
@@SynWOSall criminals breathe and drink water. TO AVOID BEING A CRIMINAL, DONT BREATHE OR DRINK WATER.
Yes, spinning is clearly spinbot weighted, but cheating bans is not something you can "mistake" with.
I'm surprised valve got scammed like that
9:45 The trick is for valve to download cheats themselves and create their own sample data from it so they can be certain that the data they are training on is actually a cheater.
You don't think they do?
@@Sommyiewell clearly they need to work out some kinks still 😂
But the dataset needs to be massive to have high accuracy rate, something that a dozen employees can't really make
you can't download all the cheats, most private cheats are by invitation
@@bruwyvn In other words - they can but they don't want to spend the money to do so.
I left the AI field few months ago. I got sick of upper management changing directions because the saw or heard someone else is doing something new or using something new. I swear in 8 months I wrote 6 lines and probably went to 150+ meetings. AI is a dead money pit. 99% of these projects are scams
A never ending battle. AI will make it interesting. Cheat devs will have to get crafty. They make huge money and give themselves backdoor access to people's stuff. They won't give up easily.
You mean 'We're just waiting for the cheaters to start using AI of their own'
Unless cheating in games can bring repercussions in real life like in South Korea, the only thing anti cheat can do is to make cheating so tedious so that the process to get cheat to work is enough to discourage most of cheaters. But this also means there will still be cheaters who spend over 100 dollars per month just to cheat.
Reminder that chatGPT was created on manually curated data.
These language prediction scripts have never been good or special.
Came for the wiggle stayed for the g0ated content, keep it up
This is why I'll never use voice chat in GTA again. They just started to use AI to scan for 'toxic' behavior.
I can only imagine how many false bans are going to be raining from heaven now
Misgender someone and you get banned
Edging rn
Hype
This was such a predictable problem.
I love how in the previous interview they say 99% and are not worried about false bans, or fingerprinting, but how am I supposed to trust that if I do a flick or micro adjustment I won’t get banned.
I have to worry about getting a false ban and cheaters get free rein. It’s unfortunate and I hope something is done.
In a way I can respect that they are trying to do something unique with the anti cheat and with the sub tick etc etc, but it’d be so much better for the health of the game and so much easier if they’d just get a damn kernel level anti cheat and 128 tick servers. Like 99% of cheaters would be done overnight if they got a good kernel level anti cheat like riot. I’d trust valve with my stuff any day way more than riot
@@meat.Valve wouldn't do that, it's too big a security risk for minimal gain
@@meat.Apparently nobody has told you this, but vanguard doesn't work.
@@meat."Man, I wish they would just install mallware on mystem and be done with it; smh my head >:("
AI also needs to learn
I think what people are forgetting is the fact that training AI cheats (the counter to AI anti-cheats) in a race against one another will be both difficult and time consuming. There may be a large market for cheats which invites some "experts" to place their claim in the market and develope AI, but generally, the experts will be working at Valve. Those people will have unmitigated access to any data or computational requirements necessary to fight this never ending war. I think the back and forth with be reduced from weeks at a time to hours. This will mean that people will have to buy new cheats every few hours. They will lose trust in the cheating system, or they will lose too much money and be unable to keep bleeding their wallet. If attrition is what cheaters want, they will be sorely mistaken.
It will just become a subscription service. But also it's significantly easier to trick and ai with another ai then it is to train the initial ai
I got banned after my g pro sensor went crazy and spun tf out of me.
I was getting a lot of "play stupid games, win stupid prizes" as if I was supposed to factor in a possibility of my hardware getting me vac banned.
💯 riiiiiight your g pro sensor "spun" 😂😂 cheating ass MF bye
Addressing cheating should be the #1 priority of publishers and developers. The gaming industry should have its own version of the Manhattan project to address this problem.
You want an absolutely perfect system for detecting cheating/rule breaking in games? Go look at iracing, community reporting, but EVERY report has to be looked at by a person. It takes longer, but Iracing support is actually the best support I’ve ever dealt with. Truly setting the bar in all cases.
Man, i don't wanna seem like an asshole here.. but you could really use a pop-filter on your Boom-Mic, y'know?
It'd completely remove all the tongue-click sounds, and would go a long way towards cutting down on the way you sound like you're gasping for air like a man on the verge of drowning between every sentence. Despite the fact that I am *100% sure* that were i in the room with you, i wouldn't be able to notice even the slightest hint of either of these two things.
Also, just wanna say; Once again, an _absolute_ banger of a video, man. I fucking love your new post-Tarkov content. Honestly, in my opinion, you've yet to miss the mark even _once_ with these more 'investigative journalism' style videos. Cheers, g0at. Glad you're sticking with it and making great pieces.
you’re a fan of ASMR aren’t you 😂
A lot of pro cheaters, what a surprise.
the cs subreddit is full of pro cheaters that hates this and loves crying "false ban" lmao
@@f3rny_66 Valve literally reverted thousands of bands because it was obviously false positive, how do u feel now?
@@f3rny_66I mean that's literally any cs forum though
bullsh*t ! The Ai anticheat just works fine. Of course all cheaters who got banned would come and say " buuuhuu i just play with 50.000 DPI ! im not cheating :("
g0at, truly the GOAT
As a software developer and AI researcher, i was heavily critical of AnyBrain pointing out most of the same, valid points that you made in this video. Bravo; great video and very detailed. Keep up the content!
They said that arena tourny wasn't actualy on LAN btw, Nikita said it was on some east coast server IIRC
Great vid! How you don't have at least a million subs and 100's of thousands of views on your vids is beyond me. Keep up the awesome content!
IMO no one needs that high of DPI, play stupid games win stupid prices... No one needs high DPI like that
Nobody ever needed more than 640Kb of RAM either
@@user-mj1vj7tb6x so you are trying to tell me that there are humans who have the reaction time to play on 9999 sens?
@@dermelonensaft3103 No
they need to NOT ban smurfing. if you've played a similar game you can be banned for knowing what you're doing.
Smurfing is cheating... Just like it' cheating for a heavy weight to fight a bantam weigh or for a Man to Pretend to be a woman and kick the shit out of women in female UFC.
Some things that you said are a bit off base:
1) Purely client-side cheats like god-mode or esp are not detectible by anti-cheat, how would they be deteced?
- This is mostly not a problem for server authoritative games. There is no god mode that a client can run if a server is the authority on who takes damage. If client 1 shoots client 2, client 2 can deny it all they want, but the server doesn't care. The server is running a copy of the game and their copy is the only one that matters
- ESP is possible when a client is an observer of an object. A client can only know something exists if they are told it exists. A client can only know where an object is if they are told where it is. Only observers are told this information and secure games manage observers dynamically and with server authority. ESP is only possible when a client observes all objects or when it is possible to observe all objects by force
- Basically, this is a game security problem, not an anti-cheat problem. The solution to games that have security problems is obviously just to learn the behaviors that players with these cheats show. As you said, if you can't get this data then it's impossible
2) The fingerprint isn't possible since the inputs can be masked
- Client side inputs don't matter if they aren't sent to the server. The finger printing should work just fine on the information that is sent to the server since most cheats aren't changing every input that is being sent to the server
- When cheats decide to mask those inputs being sent to the server, they would need to change all of their inputs. There would be no original walking or aiming input that is sent to the server at any time in this case. The user of the cheat would be able to avoid fingerprinting, but would feel that the game is less responsive since their exact inputs are being manipulated in some way. This could involve value tweaking, input delay, or false inputs
- Fingerprinting obvious can't track people that mask all of their inputs, but it's at least another line of defense since many cheat users would probably not enjoy the masking
now tarkov just needs to get its shit together
They will cash in on Arena and move on.
You have probably already considered this but I feel like taking your channel in the direction of focusing on anti-cheat in video games and doing case studies on successes/failures throughout history would be really insightful and interesting.
valve is the one company that could afford to buy a sky scraper and have it be anti cheat tower and just have 15000 employees review cases or watching suspected people in games for toggling
Lmao what kind of hypothetical is this 😂
I got so happy when I saw that you are covering this topic. :)
Not seeing the client side part..
Going way back, cod 1, cod 1 uo, we had this screenshot mechanic literally taking screenshots of the screen of the player and we banned loads of people based on just that. The same bots that did not get detected by punkbuster at the time got rekt by just reviewing the screenshots saved in a folder on your server.
I understand privacy and what not. But "rootkit/malware" kind of anticheat should be embraced if we want to get rid of these people.
Hell back in the days there were servers specifically for people who wanted to cheat.
Maybe i'm just getting old. Maybe i'm over simplifying things i don't know. But heck old internet was better.
As an ex cheater on source, it's quite difficult to be detected by you teammates or the other team.
None the less at the end it's only a game, a closet cheater who only use Wallhack can be defeat by any player.
And don't forget them the industry including valve CREATE this issue.
Before everyone was playing on private servers admins and moderators have to mange cheaters.
A server well maintain whit good staff was I sign of quality, you have to respect your teammates and play accordingly to server rules.
I got banned for spinning haven't gotten unbanned thank you for bringing light to this posted everywhere on reddit and got shamed.
On his Instagram
I always enjoy your content, because you don't sugar coat anything with these people and always ask the tough questions, good work.
They don't need to go through every single high DPI ban to determine if the ban was correct, all they need to do is have a system in place that checks if the player got a kill while spinning
So no more trick shots...
So me with my negev running into smoke and spinning is gonna get me banned...
@@blankblank8608 since u havent been banned yet it means u arent doing that regardless
@@keijokojootti7790you dont get 5 headshot in half a second with 5 bullets while spinning at the speed of mach 5 😅
@@s0meRand0m129
True.
But "...if the player got a kill while spinning" as OP wrote then none of your mentioned attributes are required.
Easiest Anti Cheat is increase the price of a game tbh.
Yeah then everyone pirates it and the game makes much less money also are you saying only poor people cheat
@@garakthetailor pirates an online game? Huh? Also no, not only poor people cheat, you can easily just get a new account and boot the game back up if you get banned on a free game. YOUR ECONOMIC CLASS HAS NOTHING TO DO WITH IF YOU DECIDE TO CHEAT IN A GAME
@@yosukehanamura3507he's saying increase the price all that does is increase levels of piracy
@@garakthetailor yeah, if a piece of software costs money, people are going to pirate it. Making an ONLINE ONLY game paid instead of free would not increase piracy levels astronomically like a singleplayer title. Also, if a game is free, people are not going to pirate it at all, no shit piracy levels would go up, but in a game like CS, the profit loss from piracy wouldn't matter because they're already shipping the game for free to begin with? Meaning they don't profit directly off installs already?
It didn't work with CSGO before and it didn't work with Tarkov
The issue is match making servers. Back in the day, we had dedicated servers only, like on Modern Warfare with communities hosting their own servers. It was easy for players to police their own servers. When they introduced match making in MW2 and didn't allow for dedicated servers the cheaters took over within a week. Match making took over for game releases around that time and the cheating issue exploded to what it is now.
Thank you console kiddies
So you kill an admin and catch a server ban.
i've heard valve made their anti cheat with the AI they use for the bots ingame. lol
It’s almost like implementing a kernel level anti cheat into a competitive fps known for its cheating issue would not only sort out false positives but detect game hacks with higher efficiency than any other method and would be hugely positive for the community. I give valve my buyer’s info, I’m not worried about them, as a private company, fucking me over via kernel access.
you shouldn't be worried about valve fucking you over. with a kernel level anticheat you essentially create another door to your system. how secure that door is is up to valve.
besides that they won't go kernel level. that would go against all the things they are planning with linux.
stuff like this is the reason why I dont play online competetive shooters anymore. my gaming experience has been amazing and much better now. the whole scene is just.. kinda a dumpster fire now anyway.
It can be frustrating how much speculation can be involved in this field, but even educated guesses can be helpful and interesting when dealing with these black boxes, so thanks!
My very recent game which was last night we had a spinbotter with an AWP on the other team just killing everyone in just 15 seconds of every round (also has rapid fire so the AWP is just like an auto-sniper). Why is VAC live not able to detect that??? :(
VAC Live only bans humans....
Closet hackers is my problem, I don’t even run into the obvious troll hacker often at all, it’s the closet cheaters, those kids with the low self esteem or w.e reason why they end up cheating. I’ve let go of competitive games. Especially games that can run on a potato like cs2
Cheating has ruined competitive gaming for me
bro, am I the only one who realizes that 99% is actually hella low? in a playerbase like CS, it would mean that 7900 ppl would be banned for nothing. wtf
On his Instagram
wait is csgo now using AI, is it finally over with cheating?
nevermnd.
ESP has very obvious patterns that an AI could detect,
like aiming straight at another player through a wall
or the percentage of successful pre aim (aiming where a player will appear in the future)
For the cheater to do any multiplayer action in a multiplayer game they have to send some input to the server so that the server can update the game state for all the other players. That is the input you analyze, the cheater can't fake this data because then they wouldn't be able to achieve their goals ingame
You're doing gods work for the competitive / pvp gaming industry. I've seen too many people stating that Anybrain would fix everything, but when I went to observe their website a year ago, and looked into the available information about them I came to the same conclusions as you have. You've of course gone a mile beyond in showcasing the bullshit they are peddling! All the glory to you, and I can only continue to follow you and see how far you can go!
anybrain reminds me of basically homelesss video from 2 years ago about waldo vision or something
The only way to feed the ai correct data is if you pay people to play on your hardware, with and without cheats and make sure they can't do anything other than play the game the way you intended them to play.
You record that gameplay and feed it to the ai. But you'd need thousands of players at different skill levels (with and without cheating) to get an accurate portrayal of the player base.
You could also choose to not ban anyone so that cheat software doesn't evolve to evade your anti cheat. Then you mark everyone that for sure is cheating with your conventional anti cheat.
Try to find each and every cheat on the market and play with it to see if the cheat gets marked.
Then train the ai on the gameplay data from all the cheaters and non cheaters. And you'll have a fully correct trained ai. Assuming people didn't get falsely marked.
Every human has blind spots, while this is only one way to "Fingerprint" this would be quite obvious, as a player detecting in specific area's of their known screen area, will show a perfect heatmap.
I am sure there is far more to this, such as when using Aim trainers, I suck at the upper half of the screen, while I a apparently pretty good at the bottom half, etc.
Hopefully this isn't just smoke being wafted gently up our behinds, Which I doubt as AI is at an extremely high level already, so it is likely they know all this basic stuff mentioned.
In regards to AI requiring humans to label examples of cheating and legit play, that is not entirely true. AI can work in an information vacuum and kinda brute force its way through the problem. The more you help it by giving it corrections the less brute forcing you have to do. The various forms of cheating ARE objectively distinguishable from legit gameplay, almost by pure definition. In other words, cheating either directly modifies the game in some way which can be detected, or in the case of cheats that "cant" be detected they modify human behavior such as having a wallhack would change how you play and behave. An AI would in theory be able to separate out these various groups without actually know which group is which and without having a single example of what a cheater is by purely being feed the replays of every single game.
At this point you would have what is basically a sorting machine that groups "like with like" at which point you can have a human come in and analyze which of these groups represents cheaters and even what kind of cheat they use. Of course, doing it this way would work but also be one of the slowest and dumbest ways to accomplish a similar thing but its more intended to demonstrate that it is possible.
Was banned from CS 1.6 two decades ago. Still dont know why and when asking they only told me that they dont know/cant tell me why and that they cant do anything. If i was cheating, i wouldnt make any sound, but till this day i still dont know why. Was it because i changed the config and had some "quick buy" options (pressing any num key to buy whole kits of gear), or because i imported some spray logos...IDK🤷♂
1:08 the stradivarius on the wall is a subtle flex
man if you would make more videos your channel would be way bigger, dont give up on something good man. You have the voice of a great story teller
This is nothing new for Valve or CS. I got a false ban 7 years ago in CS:GO, playing only casual matches and zombie escape servers, despite never cheating in a game in my life. There's been nothing I can do about it since.
On his Instagram
I completely forgot sparkles exists I watched that dude like a decade ago
That company provided you a research paper on how their tech works, but rather than try to understand it, you just seemed very condescending for them having even tried, but it just not being enough in your mind?
And why should they go more into detail for you? So you can make a quick buck and get some publicity? Pretty petty reaction to not getting a phone call. They're a tiny startup trying to do something new in the space, but it's apparently not good enough for the entitled UA-camr.
Also smurfing should be bannable, and in that case, what's the issue with the ban at the start of the video? While some bans might not be justified, being strict on suspiciously good fresh accounts like that is an extremely effective way of preventing cheaters. Those bans may only be temporary too until someone manually reviews it.
But yea smurfing just creates more work for people manually reviewing reports, so it's perfectly reasonable to be banned for this.
Also at the end when you talk about them needing to validate the clients input came from their hardware directly, that's not possible. You receive data from the client and validate it from the server, no software on the client will ensure it hasn't been tampered with. Never trust the client, it's just a rule. Solutions like TPM will help prevent tampering though, but most games still don't enforce this as strict as they can because it would limit their player base too much.
You know when you feed AI an image and it will throw you what it thinks it is and the percent confidence it has in that prediction? Say Duck (99.7%)?
The fact that the AI can't be completely confident, as well as the fact we can see it being completely wrong sometimes, should basically kill the idea of relying on AI as the arbiter of cheating here and now.
AI will get better and is a good way to expand the net and find suspicious people, but we need humans reviewers to at least look at appeals and hand out final decisions.
I mean if they linked credit cards or IDs to accounts and banned that info so bot and alt accounts cant happen as easily that would fix so much
People will say that cheating is an issue, and that the solution is for the devs to make infrastructure to stop it, but time and time again, we see that devs and players both working to stop it is the best solution. CSGO’s Overwatch was legitimately useful, the issue was that cheaters figured out how to game the system. Better bot protection and a more open platform for the community to examine the data themselves would prevent obvious flaws that existed previously from going unnoticed, hell, they could go hand in hand. I think the use of AI is definitely a good idea, in so much as catching obvious cheaters quickly, but they do need to introduce some human element into it, especially considering it has the ability to VAC ban. Perhaps even an Overwatch LIVE system could even be implemented, to give CS2 streamers and players something else to do. Obviously it can’t be 100% real time, but spectators of competitive already have about a 1 round delay anyways, so they could easily just import that. While I think Valve should definitely be utilizing AI to empower themselves, they should also be utilizing it to power the community as they did in the past. It’s also strange seeing Valve move from VAC bans having a near 0 false positive rate to a large number of them being false positives as of recent. Definitely represents a mood and/or ideological change in the development team imo.
overwatch had it's issues.
from 2015 to 2018 you needed 11 unique reports to have your recent match(es) be sent to overwatch. "a game a day keeps overwatch away" was the motto.
from 2018 onwards they implemented some sort of scoring system to automatically send matches to overwatch. something a long the lines of -> a few games with ~100 score and you're in overwatch.
in 2020 i think? they started to automatically send rage-aimbotters into overwatch. this is where overwatch lost its purpose. it was filled to the brim with cases that could have been handled by software therefore making room for cases that require actual human intervention aka the closet-cheaters.
i agree that overwatch isn't a bad idea. but the conditions to get sent into overwatch needs some refining and it's near useless when all you get is ragecheater after ragecheater.
7:20 I wonder why with all the layoffs and them not hiring
8:17 pretty sure if we used 150th of the military budget. Lots and lots of cheaters would be gone
2:46 I got banned 15 times by Overwatch without cheating in any of my matches in times when you had to pay to play cs:go...
Nice video, thanks for the content.
I just want to point out a mistake you make on my point of view: you state that training AI with already identified cheaters will only allows to detect the same kind of cheaters. BUT: you forgot about the fact that if a human use let's say 4 variables to identify a cheater, when the game registers 12 variables. If you train the AI giving it the whole dataset (so 12 variables), then afterwords it will be able to identify cheaters based on the others variables that the human can't see/handle.
A good comparison would be AI trained to detect diseases with cameras. We use hyperspectral camera that catches more or less the whole light spectrum. Then a human identifies the disease on each picture, but the human only uses its eyes, so only a portion of the light spectrum. But the AI now know that the other spectrums of light are like this or this when there's a disease.
So in the end, AI will help to catch knew type of cheaters, if developped properly :)
So strange to attempt to solve this incentive problem without modifying incentives. The gaming industries should be campaigning to make being a cheater in video games be a real world crime with real world consequences: fines, community service, jail time; this also benefits from the requirement of enforcement being as delicate as it is in real life, beyond any reasonable doubt with unions and an appeal process.
Nah bro that’s wild. Any kind of policy put in place would be abused and be 10x worse than DMCA in terms of suppressing creativity and consumer right.
This is an interesting point, and I wonder if it likely already is a crime since you're using a company's network and infrastructure in a way that bypasses their security.
Cheating is Illegal in China... The country with the most cheaters... You can't rehabilitate cheaters... Once a cheater always a cheater and cheaters reproduce with people creating more cheaters and cheaters reproducing together creating even stronger cheaters... It's a virus...
While I'm absolutely for punishing cheaters with fines, jail time and so on, this idea is very very sippery slope. While it should be illegal to tamper with software to gain an advantage over another person in an online game, I'm certain that the fucked up gaming industry giants would also try to outlaw modding as well. We've seen companies try this shit in the past (rockstar for example). This would bascially kill any unoffical modding which is a huge loss for the industry.
They're trying to make something Easy more complicated.
Secure clients are not really ever going to be an actual solution unless you have to go to a physical location with security and human surveillance, show ID, and play a game on the company’s computer like how we do professional testing.
Sever-side validation will remain the primary mechanism to validate user input. Checking that humans are pressing physical keys isn’t really the point anyway, for example disabled gamers may need virtual inputs or whatever.
Cheating is a social problem and thus does not have a technological solution imo. Humans have cheated long before games were on computers and we will continue to do so. I think chains of trust (I.e. you vouch for me and I vouch for my friend) are a promising way to find groups of players that you want to play with and it can be automated and scaled with more technology. That may not work well for competitive games though. But to be honest we can’t even stop people cheating in physical sports either so 🤷♀️
While you provide interesting feedback and views on the topic and this product, however I believe you're too much stressing about the client side here. While I agree that the communication from Anybrain could be better and a little more transparent, I think you're simplifying this topic a bit too aggressively. Additionally, less is literally more for AC developers.
For example, this kind of AI-based detection is solely based on human-factor. It doesn't need to know how ESP looks like for the user or what inputs the user gives, but it needs to distinguish the behavioral patterns between different cheat features and legimitate users. And as you said, this requires a great amount of training material that has to solely consist of already known cheaters (& cheat features) and non-cheaters.
However! This doesn't make it impossible, but it does mean that an AI-based anti-cheat detection system WILL HAVE TO work tightly together with traditional anti-cheat detection tools to get the required training material. Over time any AC client will be able to gather the required data, as long as the AC developers are conscious of the technical specification requirements of training material.
There are already a lot of studies available from other fields that can be utilized here. I'm sure this will be only a matter of time until we've a working solution available. Perhaps Anybrain will fail, or perhaps not, but there will be something similar available in near-future.
Bring in pros to the building, closed hardware, use them as a baseline, and ban the above. If a false ban is reported, they must prove it, and if able they become part of or are the new baseline.
Same for hacks, use actual hacks on closed hardware so the system knows what its looking for when they are used.
Human oversight will always be needed in the next decade.
I've played with Circuit players, they can easily play like regular people when they aren't playing for positions.
Your argument relies on professionals being incapable of playing like players, which is not only insulting but most importantly ludicrous. @guymontag5084
Thank you for bringing this new teck to light and explaining how it works
If real, I would assume their looking at the data sent from the client. While the data could be doctored, it would probably show signs of tampering. (Much like photoshopping a photo.) They could be looking at other factors, like latency between sent vs resaved data. Variations in the data. Average steps in the data.
While it may seem like an impossible task. This is the same issue people face with virus, malwares, extra.
Would making a cap on dpi help Ai catch cheaters?
overwatch was probably the best anticheat you could hope for
Keep up the great work!
I am going to get pitchforked here, but I am going to write this wall of text anyway, maybe someone can see through the blinders because of it. If one person would, that makes it worth my time. I am pretty sure that pro players in CS GO were cheating on all LAN events in past years. Valve did absolutely nothing to stop it, because they do not care about integrity of game as long as it is making them obscene amount of money. They let kids gamble, let people ruin their lives for pixels. Goat already touched this subject in 2 part video series, BRAVO! Nothing else matters for them.
I believe Valve FINALLY found solution to cheating problem in Counter Strike. The game is the CHEATER now and it doesnt matter that Billy has walls or aimbot in grand schema of thing because it just doesnt matter anything in result of match. Bare with me here:
1. AI powered driven big data matchmaking
2. AI analyzed chat and voice chat in games to see how players react in match by game cheating on them or they accuse their opponents of cheating - likeliness of player quitting game session
3. AI controls interpolation (peeker's advantage), crosshair movement (micro adjusts your shots to hit or miss opponent), change recoil patters, hearing range for footsteps, flash bang effect duration,... I have seen all in my games and there are quite a bit more in there.
4. every player in game has good and bad moments within each match or at least in series of matches to make him "feel" good at the game to continue playing
5. cheaters are irrelevant because game can "make" them die by legit player just by manipulating game itself = cheating problem solved?
6. players are unaware of their own true skill and basically spoon fed dopamine and rank without progressing in game as they should - if you play CS for 18+ you just know
Feel free to comment about your game experiences. How do you like new CS2 compared to CS GO, CSS, CS 1.6 or even older version. I started with CS 1.0 and I am pretty sure this one is by far the worst CS ever release.
nah you got to take your meds bro, you're living in 1984
as someone that's been making hacks since like 2000 I laughed so hard at that 99% naw man you can 100% keep hackers out of games now they just don't because they need someone to sell the game to after it's past it's prime. Same with botters. Look at blizz they print money every 6 months. Why 6 months? Because it gives botters time to buy back in and work for 6 months. Hell the turn around rate on blizz bots is 100% because it's to good of money for both sides to pass up. Just think you can sell a game to a hacker at lest 4-6+ times with a botter they will always turn around and buy back in. It will never stop you'd be dumb as hell to 100% stop hackers and botters in your game.
When i play 1.6,source or CZ it feels like cheating when I play 2 everyone just camps corners
Nah, man, no apology needed. This is the actual important stuff to discuss.
I will say, g0at, that synthetic data is the long term solution to high quality data. Maybe we don’t know what a cheater is from Overwatch, but, if we can successfully synthesize counterstrike games of sufficient quality (or images of ducks, or well-argued essays, whatever), then we can significantly increase the accuracy of these systems to distinguish between players and machines. But, yeah, obviously if anticheat datasets can synthesize realistic human gameplay with machines, so can users.
My solution to avoid cheaters: Stopped playing pvp games
If I had learned anything from my machine learning class in college, machine learning can find things humans can't but only if the algorithm and the training set is suitable. A poisoned dataset is normal for these kinds of problems for the AI, and that the model itself should be able to figure out. If it doesn't work, its either a bad algorithm/implementation , dataset, or training power.
@8:45 AI be sounding like police in the US. 😂😅
Valve is all about automation and doing the least amount of work for the most profit. yeah this seem right.
yooo i was just rewatching some of your old videos today hoping you would upload something really soon and you did :D
There is a simple solution to reduce cheating.
Simply dont allow russan and chinese players to play on EU servers.
it would still be nice to have a basic anticheat that will ban obvious cheaters quickly and effectively, not requiring a manual review or player admin etc
Good take on AI - thanks for presenting it as a tool of many and not the one to rule them all
The reason they can't show proof of their 99% claim is because it's complete nonsense.
Feels like the start of a black mirror episode
Ah man I can’t make another account and start cheating right away