Notices by anonymous (anonymous@freespeechextremist.com), page 19

@creamqueen @Zerglingman Some of us like to not waste any space. I know that’s a foreign concept these days though. :alexjonessmile2:

@Zerglingman @ceo_of_monoeye_dating @lonelyowl I know. If my tinkering yields positive results obviously I’ll go with that, but I just don’t see it being a particularly viable root in terms of mass adoption.

I was considering a mixed approach though depending on those needs. Like perhaps using ML/RL for using attacks and what’s the best attack to use.

I was already struggling with how to make movement not make the game slow to a crawl in the first place, so perhaps it’s all for the best. :yuiShrug:

And if you’re interested, I might whip something up for the fun of it just as more of a “what if”.

@ceo_of_monoeye_dating You know, on second thought, perhaps I shouldn’t do it at all. Don’t want to waste the brain space and fail. I just wish I had some benchmark or something to work off of.

I don’t know. I’ll give it some thought while I get everything downloaded and ready. In the end I might just take your advice and scrap the idea altogether.

@ceo_of_monoeye_dating I have access to the underlying database in this case, so it’s getting access to raw game data.

>you got this 5 times for a reason are you saying you TL raped me ON PURPOSE?! :alexjonesdemons:

@ceo_of_monoeye_dating I’m not deadset really. If it is that bad of an idea, regardless of how little it needs to process at any one time, it would be far better to simply make a traditional bot. I was just wondering, given how little I expect it to need access to and how little it needs to ultimately do, if it was still a terrible idea.

Seems it is.

MDK is markov chain isn’t it? Q-Learning. I was also thinking of going observational but that’s even more intense so doubtful it’s a good idea.

There was a program for RL Q-Learning which used something called sweep learning, and supposedly it’s much faster as the expense of static resources like memory (at least that was my understanding at the time).

Any idea if any of these newer things are better or worse?

@ceo_of_monoeye_dating Alright. Well I figure I’ve put in the time learning it, might as well at least try it. Since neither of us knows just how bad it will ultimately be. Shouldn’t take me very long to implement.

I have a pretty puny computer when it comes to computational power, so if it doesn’t work for me likely wouldn’t work for others.

And worse case I just get started on a traditional bot. Having access to raw game data should make my bot relatively easy to make, at least a dumb bot.

This will also let me peer behind the looking glass of ML, hopefully learn something.

All in all a good learning experience if nothing else.

@ceo_of_monoeye_dating Came across an article from 2020 >https://probablydance.com/2020/01/29/why-video-game-ai-does-not-use-machine-learning/ points 2 and 3 already worry me, and given the advice I’ve gotten from others that work in the field from this thread, it’s looking like less and less of a good idea to go through with AI. I would be interested to know if this has changed in 3 years, so if you know anything please let me know.

In the meantime, even if I can get around computational issues you still have issues with needing new models for everything.

It would certainly be interesting to see if making the bot “observational” changed anything, but we’re still talking issues that have yet to be addressed.

The more I read into it the more I change my mind.

@ceo_of_monoeye_dating I got this 5 times for some reason.

What if the underlying database is relatively small (lets say <100MB) and the operations rather minimal (like do thing record thing repeat). Is that still really bad?

Like no real time bot training. More like train-then-use.

@lonelyowl @Zerglingman IDK fruits are pretty sexy :1000:

@supersid333 @Zerglingman Yeah OpenAI is totally cucked. The model I settled on now is not owned by OpenAI as far as I can tell, but it shouldn’t be too hard to make a switch if I have to.

@Zerglingman @lonelyowl >labeling fruit as gore :alexjonesrarrr:

@lonelyowl @Zerglingman >no labeled data of corpses and blood prudes

@lonelyowl @Zerglingman What keywords did you use to find that by the way?

@lonelyowl @Zerglingman >The full dataset after compression is 365 GB :jokersurprise:

I have to be honest though if I had the space for that I’d totally use it for custom maps.

@Zerglingman @lonelyowl I suppose I disagree. Not so much on the base premise but that it’s not always the case. For example there’s a single player to StarCraft, but the multiplayer might as well be a completely different game. However for StarCraft at least you can add bots to the game, as a stand in for players. Which at least implies both were meant for people to play by themselves at least in theory.

I could just be tired though. :blobsleepless:

@Zerglingman @lonelyowl Well I am not sure what you mean but I think you mean these are all games you play against another person or bot? I suppose in my head I was thinking PvB = PvE. In any case games like Diablo would still benefit, and shooters as well. It could also lead the way to more realistic environments and mods in games in general.

@Zerglingman @lonelyowl In essence, I am suggesting what RL models already do, but for much larger and more complex games. Something no one has been willing to dip their toes into yet (far as I can tell) probably because of copyright™ concerns.

My ultimate goal I suppose is to reduce burdens on mod makers while also disincentivizing always-online bullshit. Maybe this is a bit too much power to use, again I haven’t even gotten started yet, but I still think a goal worth striving towards.

@Zerglingman @lonelyowl I don’t understand :acelevy:

@Zerglingman @lonelyowl ??