wife stolen dont click

qw already has pretty tough bots doesnt it

frogbot 100 (or slightly less even) are pretty hard to beat!

Why cant we have that kind of bots in quakelive. The cpma bots went up to 100 level as well. Was fucking rediculous how much better it was!!!

but they have routes and manual programmation,
with these projects you are more likely to get humanized bots

misguiding headlines here, and on slashdot (who the fuck uses slashdot in 2016)

interesting concept of a bot tho

misguiding headlines?
Which sites are better than slashdot in 2016? :-)
I sometimes still read bluesnews ;-),

The headline totally misses the point... : it's trivial to write a perfect bot that hooks into the game's internals and always wins. It's difficult, and more generally applicable to make a bot that learns to play by watching only the same info the human players get: the screen buffer.

bring it on

Is it possible to add cpma AI to quakelive through the workshop?

please god ill donate some moneys if someone makes bots like cpma to ql..... Just speen like 18 hours past two days playing vs 12 bots on CA Castle DeathSTalker trying to make it more challenging,

Get 12 persons for each team, start, then individually delete my team players, so im playing verses all 12 at once. crazy chases etc.

Vapirc version of this is crazy too..


Just wish i didnt have to do stuff like that to make it interesting lol

[IMG]http://img4.imagetitan.com/img4/small/14/14_350kdamage.png[/IMG]



Oh well cant get the picture to show up but its clickable.

Hehe, I see I'm not the only one making custom challenges because the bots are boring :p

One I did a while ago was try to win a duel against Major (Nightmare difficulty, normal think time) on Cure by a 35 frags difference. You can change 35 by another number to match your skill. But once it's setup so that you have to rush her a little more than you'd like, you'll rage at when she plus-backs with her defensive rockets :p
The challenge is first to be able to do it, then to be able to do it twice in a row.

The cpma AI (Apheleon lvl 100) you are talking about was a modified version of spiterbot created by some Russian (Rylka or Ryalka something like that). Spiter bots or Spitebots had strange movement, not strafe jumping but actually doing giant leaps and were aimbotting (in the sense they would always hit 1st when coming out of the teleport on t4, i.e.), 100% MG and so on, there was massive cheating involved, not some kind of human like bot.

I played spitebot for about 6 months back then and I did not improve at all because in order to beat the bot you only develop two attributes and learn one thing: you learned his patterns and you develop reaction time and the habit to hide, to keep your playermodel always behind corners so the superhuman aim won't harm you. I remember that I was pissed off when I realized that the 6 months did not help me one bit to get better at q3, having the opportunity to play a friend of mine who did not play the game all that time, I was expecting to destroy him as we quite equal and we basically had the same games like before.

Playing cheating bots who would force you to play and use strategies and angles you don't use against human opponents is a waste of time so cmpa bots were shit.

Depends what you are training. I just want to practise timing and memory with a bot that will time and fight for items. You can't do it in QL because the laughably named "Nightmare" bots don't even bother so you end up taking everything on the map all game so it doesn't train you at all. It's all about being under pressure from your opponent and trying to remember the numbers in your head because that shit is hard (Well as least for me). So having a bot that times items and strafe jumps around is all I'm really asking for, cannot believe how terrible QL bots are urgh.

You could not even train your timing with those bots because they would also time and come on time and thanks to their cheating aim you could not put yourself in advantageous positions -like between two powerups or to delay his pickup -because he would just enter and destroy you while picking up the items like you would not even exist. Thanks to this you could not build up cohesive play around timing either. I agree that QL bots are very weak but the Realka bots are no better.

Well you just set the accuracy lower for them. I don't need to physically pick up the item as long as im in ears shot every time. Doesn't even matter if I die because its just a bot and the memory is the main thing i need to improve.

with respect to spiterbot's movement -- from what I remember, the official explanation was that they took an average movement speed of pro quake players and made him move with the same avg speed. but since he also walks pretty much, his strafe jumps are much faster/longer than normal.
it looked really weird. sometimes I created a match between him and anarki and watched his pov. e.g. I had a demo where he makes one jump b2r shooting in the same time anarki (so he looked in a different direction and the jump was done completely sideways) with about 100% mg and right after touching the rail platform he immediately jumped back to bridge without releasing his aim lock even for a moment.
also there were a lot of funny moments when he aimed someone during one frame and the next frame he looked to the previous direction. so it looked like he's running somewhere, then you see a short 2 frames long blink and nothing more changes in his viewing direction, but you see the message that he railed someone.

It's not just Apheleon 100, there's also Rat and arQon. Besides that, any bot can be added with 100 skill.
The bots were made by Qrealka.

Their aim is really good because that's how they were specified in the bot files, though you could actually find another bot that hit a bit more human like accuracies (which are not actually 100%, but not far off).

But yeah, basically cheating movement wise, very hard to predict at times.

Nub AI shitty movement

as much as the idea of on screen learning is a nice improvement. I somehow doubt the bot faced any proper player.

afterall, even the shittiest ai bot will have good aim which a noob cant compensate against no matter how much item control or traps he sets up

The thing is that AI can practice 24/7 on zillion threads simultaneously, so with enough time, you don't really want to face such AI in the arena, let alone in real life.

But this is not chess, where a computer can calculate the optimal next move for a minute or so. At least for now it would have to rely on some kind of heuristics too.

But you can calculate the optimal move, quake plays like a game of poker so you make a bot that keeps tracks of all armours and hp and enemy weapons/ammo/stack (by making sure to be in hearing range) then base decision making off what is spawning next and the difference between stacks and you calculate odds based off positioning and accuracies. If you designed a bot that played this way you could have its accuracies fairly low yet be effective at duel.

I think (I guess) you are severely underestimating the amount of progression AI went through in the last decade.

My point was more that no matter how big your virtual brain is, it is still dependent on the input it receives. The AI that beat the wordclass GO player had insight into all variables of the playfield all the time. That's not possible in Quake.

Or as aeoL said: "a bot that keeps tracks of all armours and hp and enemy weapons/ammo/stack (by making sure to be in hearing range)"

And that's already a very optimistic statement, being in hearing range constantly to track all item pickups.

One could say that due to the randomness and the uncertainty the skill ceiling is not high enough in Quake for an AI to truly shine and to become unbeatable.

I already lost you at the assumption that an AI would need to "hear" an item pickup to guess what the enemy has. That's definitely not how learning works. An AI is not some dumb pre-coded calculator program, it will simply learn the map and what people do in any given area/situation. They could even feed it all the previous demos/matches, and it will also learn from those too.

The demos it learns from will be from different players and styles.

It will be an approximation. About the same worth as looking over Rapha's shoulder before you play him, and then thinking you've got his number.

The things which make you win at these games are VERY subtle at the top level, its not going to be possible for A.I. to pin it down to any definable trait and exploit.

The game is too dynamic and humans will simply inject things into their play to throw it off, the same way we slow down in cars when we know there is a speed camera, then speed up straight after. Or act nice when we want something!

Quake is kinda like a 3D version of chess when you consider unlike a chess board, a new map change means a whole new degree of learning again, that's why so many people only play a few maps. Its simply too big an investment to make and they don't like raping you on one map, and losing on another.

So I don't see the A.I. doing it. At best it will simply be a more consistent performer because unlike people it won't lack motivation, or have a life to get in its way, or be tired etc.

But without cheating, I don't see it happening anytime soon.

Learning and applying are not the same thing. An AI can learn to select the best move (read: the move that on average gives the best outcome) in a certain situation based on certain parameters (enemy status, own status next item spawns etc.).

In practice it does not have insight into all parameters. And there is also randomness, where will the enemy spawn after I kill him and will he go left or right? No matter how well an AI is trained, it still has to make (educated) guesses if visual or audio cues are missing.

That's exactly what I was thinking about, when some part of the AI is making educated guesses and weights in its findings into the final resolution (think about Google's RankBrain for example, it works amazingly well imo)

That does not make it unbeatable then. Especially as a good AI would also sacrifice 'best decisions' in favor of creativity to some degree, otherwise it becomes too easy to predict itself.

I agree, but I didn't say it will be unbeatable, just "you don't really want to face such AI in the arena", implying that it could be very a "deadly" bot.

I'm yet to see a video of what title says. Anything I saw was one AI won against another, or AI lost to human player albeit closely.

AI beats noobs at videogames, just like it has been doing since 1995

Somewhere in Helsinki there is a god named Milton laughing his slender asscheeks off

Milton has responded to Slashdots bot in a video

was zyzzz a great player?

Yes, he was a killer CPM player. We called him the flyingwombat. you can Ask newborn for proof.

Ok im asking *gets police voice enhancers*


NEWBORNNNNN!!!!!!

Where is u be, to tell to me, all my wondrous things, which i cannot contrieeeeve, untill the darkeness of eve, untill my bellows release,beneath the scarlet hue, the deserted wastelands, of dee-dest-dreality!

This is way more impressive than a typical aimbot. If this AI really learned from vision, then it can learn any other video game. After it's training is complete, it can be applied to real autonomous weapon systems.

Skynet is becoming real.

Everything has a certain amount it can learn when it comes from nothing and appear competent.

But the devil is in the details, and that's where you need exponential levels of improvement which is most likely not there in this sort of game.

I guarantee I will beat any bit of AI on the planet in any FPS or fighthing game with nothing more than 24 hours notice of that game.
(Providing the A.I. does not cheat in the amount of damage it delivers, rate of fire, reload time, health.

If my fireball does 5%, so should theirs.
If my railgun does 25%, so should theirs.
If my shotty takes .4 seconds to reload so should theirs.
if my health bar is 100, so should theirs.

Even if they are allowed to cheat with 100% accuracy & 100% reaction time and have wall hacks so they know where I am 100% of the time (FPS), they don't have the ability to win without cheating on core aspects, which would render them cheats and defeat the whole point.

When I can no longer make such a claim while being relatively inactive in game, then I'll get excited :)

For now, this is just something which only sounds good when the readers are clueless.

You are not impressed because the AI doesn't play like rapha. It is easy for a human with years of gaming experience to not be impressed by an AI that has only been playing for a month. It is easy to mistake the difference between a self-learning AI player and a pre-programmed aimbot. When an AI exists inside the computer, gamers will be prone to say that it is cheating. Therefore, we must build a robot, with arms and eyes and no direct connection to the desktop in order to convince the gamers that it's victories are fair and square. For the computer scientists that do have an understanding how it works, such a robot is not necessary to invoke astonishment.

"You are not impressed because the AI doesn't play like rapha."

If i said Rapha I would have used him as an example. Didn't read anythign else as I can tell this is going to be one of those times when you're putting words into my mouth that i never said to suit your opinion.

What I said was clear and concise enough for anyone to follow.

Apparently Starcraft II is next: https://deepmind.com/blog/deepmind-and-blizza...vironment/

It's from the guys who developed the GO AI which eventually was the first AI to beat a world class player in that game. However, for the reasons discussed in this thread (uncertainty, having to make guesses etc) I doubt it will be able to challenge the top Starcraft players.

If you think starcraft is harder or has more uncertainity/possible outcomes in the strategy layer than go then you are gravely mistaken (at least in the practical, not game-theoretic meaning).

Chess, Go, Jeopardy were clean challenges, but starcraft is not because it has mechanical aspect. If they are going to mechanically handicap the AI (I think there is consensus you can "break" the game with perfect mechanics just like you can break quake with 100% aimbot), whatever the outcome it will not be definitive.

Anyway, the real question is: what race is it gonna play? :D

I don't think that the mechanical aspect is the crucial difference here. The GO AI was able to make use of immense search trees to plan ahead. A Starcraft AI will have to work on different levels. First, the highest level which determines the main strategy (which I agree, is less complex than in GO). From that follows a build order for buildings and units and how to optimally harvest resources for that objective. And in the end the AI must control every single unit to optimally serve the higher levels.

While I think that an AI will be able to master the strategy level if you only feed it with enough demos to learn from, I have my doubts about it being flexible enough to handle the lower levels better than a human player. Those people at deepmind know that too and said that we should not hope for it to beat top players, but I am excited about how it will work out.

It will play all races as Blizzard plans to use it to optimize the balance between races.