ECHO, the 3rd person action\puzzle game was a fun concept to script in your machine dopplegangers to learn on you (and repeat after you one of the set actions you can do) and reset every cycle.
I don’t think it would work by itself without such limiting.
I always got the impression it wasn’t a learning AI but rather a very limited “Has the player pressed the run button? if YES: AI can use run next cycle”
Yes it is, it’s 100% scripted. And yes, in the environment where you can do like 10 different actions, they start to do their routine adding ones that you used in that cycle before they get reset. In a sense, they act no more natural than monsters from a tabletop game.
But these do make me think that if we talk gamedesign with a LLM as an actor, it should too have a very tight set of options around it to effectively learn. The ideal situation is something simplistic, like Google’s dino jumper where the target is getting as far as it can by recognising a barrier and jumping at the right time.
But when things get not that trivial, like when in CS 1.6 we have a choice to plant a bomb or kill all CTs, it needs a lot of learning to decide what of these two options is statistically right at any moment. And it needs to do this while having a choice of guns, a neverending branching tree of routes to take, tactics to use, and how to coexist with it’s teammates. And with growing complexity it’s hard to make sure that it’s guided right.
Imagine you have thousands of parameters from it playing one year straight to lose and to win. And you need to add weight to parameters that do affect it’s chance to win while it keeps learning. It’s more of a task than writing a believable bot, that is already dificult.
And the way ECHO fakes it… makes it less of a headache. Because if you limit possible options to the point close to Google’s dino, you can establish a firm grasp on teaching the LLM how to behave in a bunch of pre-defined situations.
And if you won’t, it’s probably easier to ‘fake it’ like ECHO or F.E.A.R. does giving a player an impression of AI when it’s just a complicated scri orchestrating the spectacle.
For most games, it's not difficult to make AI that can absolutely destroy humans. But it turns out to be very difficult to make AI that feels like a fun and engaging challenge to a human. Hardest of all is making AI that realistically plays like a human does.
I’m playing the PC version of SMCP, and the only difference I can notice, maybe due to the better hardware, is that the game seems to be a bit faster on PC than on PS2. And have yet to test any of the other collections Sega made for/with the Sonic games.
The Rain World Animation Process.
While the title suggests only animation, the AI is tied directly into the animations so you gat a 2 for 1 deal in this video.
I second Deep Rock. I’ve recently got back into Destiny 2, which could be an option too. But it’s also a lot more than a simple fps. Plus it’s infuriating and addictive.
I used to be a big Destiny 2 player, but my friends (UK) and I (US) could never line up timezones for raids etc. so I used to just solo patrols and got bored with the rarity grind. I did hear the last expansion was excellent but also fuck Bungie for greedy practices. I was excited for Marathon, but that’s dropped a little.
Yup PVE only. It can be as chill or as hectic as you want depending on the difficulty you choose. Lots of upgrades, 4 unique classes, plenty of mission types, good solo or team gameplay.
This has been discussed a lot over the decades (with some VERY good articles written by assholes we try to pretend don’t exist))
The gist of it is: AI cheats because the alternative isn’t “fun” and rapidly outpaces humans.
Because in an RTS? After you get a build order down, the big decider is Actions Per Minute (APM). From a build standpoint, it is the idea of triggering the appropriate research the absolute second you have enough minerals. From a combat standpoint, it is rapidly issuing move and attack orders so that you always win the combat triangle. The former isn’t significantly different than just having cheaper research or faster build times. The latter is actively demoralizing in the same way that we all died inside when we first got permission to go online in Starcraft. Except at a level that even the good players realize they ain’t shit.
For grand strategy games (barring real-ish time ones like Stellaris) you basically have two real approaches. The first is the games with research options (… like Stellaris. Look, I have been playing a lot of Stellaris lately). We try not to acknowledge it but RNG has a massive impact on that when you really want to get torpedoes but no options are popping so you are just doing the fastest research choices you can to get a new pool. And the difficulty option there is… a known order.
The other are the very elaborate fixed tech trees. Obviously this gets back to build order. And the reality is… the benefit gained from rapidly updating the hard mode AI to use the current meta just isn’t worth it. That IS somewhere that an optimizing function can be applied to (and… semi-off-the-record but that has been a thing for over a decade and is why devs aren’t THAT surprised when a “new” meta takes over in a strategy game) but it becomes a question of how much it is worth it.
All that said, we are seeing a lot more effort put into “learning” AI in racing games (driveatars) and fighting games because those tend to be cases where even the best AI is still expected to be “human” and we aren’t TOO demoralized when we realize we are in a pub with Daigo. That said… there is a reason that modern SNK Bosses tend to have super armor rather than frame perfect inputs. Because the former is “bullshit” but the latter is just mean.
APM actually does jack shit. You can spam a button fast and you’ll get 400 APM and get rolled by someone who does 40. EAPM is where it is at. Which is effective APM. How many actions you can do that move you closer to victory. Instead of just spamming two buttons on repeat (which is what a lot of Starcraft players do)
There used to be AI’s integrated into Starcraft 2 and later actually playing the game (like a player would) online. You can put restrictions on eAPM for these bots. You can force them to make human mistakes - delaying upgrades. They can get pretty well aproximated to human skill. The main issue with it is they suck at context. They can’t really “remember” stuff happening. Picked up a dropship and it flew away from my FOV? It’s gone. Oh shit a dropship came from the exact same spot! Oh good it flew away, which means it can’t hurt me no more.
There are also tournaments in SC2 for unlimited AIs - where they play the game without any caps. The only thing that matters is who wrote a more efficient bot. Machine learning isn’t reallly used there, more likely a decision tree. Those do exactly what you are describing. Playing against those as a human is pointless and would get someone who introduced them as a difficulty instantly fired.
Yeah this might be the direction I go, I have Titanfall 2 on Steam and played a little on the Deck and enjoyed the movement. Any suggestions as a N00B and any things to avoid?
Definitely toned down from Titanfall, but if you liked that game you’ll probably like Apex too!
I would say stress about the character selection too much, you’ll find characters that you enjoy more over time but abilities don’t overshadow good gunplay in the game.
A few small tips:
heal your shields before your health (most of the time). Shield items are slightly faster to use than their HP equivalents but provide the same amount of healing (e.g. shield cell = 25shields ~2 seconds to use, health pack = 25hp ~3 seconds to use)
when you kill an enemy and they drop their death box, you can loot their shield from the box to instantly recharge some or all of your shields. It can be really helpful even in the middle of a fight or if you know another squad is coming to third-party you
don’t get too caught up looting, you’ll get good stuff a lot more quickly by taking out other teams who have already looted for you
always be aware of where your squad mates are and make sure you’re close enough that you can get over and help them if need be
Best way to learn ofc is to just jump in and play some matches. Have fun!
Yeah, but it’s not like an LLM or nueralnet thing. The kind of AI used for video games doesn’t need all that to feel smarter/harder.
Making a bot harder is actually easier than making it easier. It’s super straightforward to make one that always wins and is perfect. It’s more involved making a bot that doesn’t always take the best path or the most efficient way of completing the thing. I, personally, could make a Rocket League bot that plays the game better than any human since it’s all just math, and the computer is a calculator. I don’t think I would be able to make one a human player could actually beat though.
L4D has a mechanic like what the OP wants. It’s just not very good (IMO it overcorrects way too hard in both directions). Every time you win or lose or this happens too much, and this happens too little, it keeps track of that and then just adjusts things to change it up. Like if you sprint through one stage without resistance, the next stage will have more infected to deal with. They even gave it a name: the AI Director.
The alien in Alien Isolation is like that; but it is better done.
I played the original back in the day, although not as far as the remake. It’s incredibly faithful to it, almost beat-to-beat, but carefully polished and improved in all the right places, e.g. by improving pacing, creating a more interconnected game world and subtle balancing changes. The nostalgia is still there, but it feels modern nonetheless, not just because of its outstanding presentation.
I actually don’t like horror and jump scare games. The fact that this game managed to draw me in and keep me playing is quite the achievement. This is one of those games that is so good, it’s enjoyable even if you don’t like the genre.
bin.pol.social
Najstarsze