I am always interested by these “its so simple” solutions that no one in the entire industry is doing.
I guess every single game developer is just stupid.
3 Likes
Game development and ML engineer don’t usually intersect
It’s simple for an ML engineer, but if you read up a few posts of mine, that field has changed dramatically the last few years
Once again. Just saying I find it interesting when forum posters suggest something that would be “totally so simple to do” that absolutely no one in the entire world is doing.
Like I said.
It really just means every single video game developer in the world is stupid.
1 Like
Would you like me to recount my experience with genetic algorithms? Or maybe you’d like me to talk about the ins-and-outs of generative ai, the model things such as ChatGPT is based on? Whats your preferred tool for Neural Network generation, and what sort of weighting algorithm do you use to balance out the interactions between the different neural layers?
I’m sure that Blissard has been looking into this for a while now. 
Because that is how they justify their jobs, some of them?
1 Like
yall ever add bonus stats to simcraft to see how you’re going to perform next tier for that preemptive bone?
I don’t know whether to roll my eyes or laugh or both. I imagine 30 Blizzard locked in a room for months trying to come up solutions to balancing classes, and then along comes an epiphany in the form of a forum poster and light bulbs go on around the room with the intensity of a million white dwarf stars.
3 Likes
Like when the world’s best Go player lost to AI, then the AI was beat by an amateur with a simple strategy…
https://youtu.be/l7tWoPk25yU?t=315Until AI understands what it’s looking at, there’s no point in investing them when they can be easily exploited.
1 Like
You do realize it’s a pretty big decision? You’re completely ignoring like the facts we’ve stated earlier:
- ML engineers cost a lot since there’s few of them and the value they bring is large
- until recently, training ML model was very expensive because of the need for sophisticated hardware
- Developers worry that they will be lose their job because of it
- It would be a large project to invest time in and that would divert resources from other things
- Wow is in maintained decline so they don’t want to invest in technology like this without a good reason
- Misinformation and lack of understanding (as is apparent by the majority of people on this thread) of AI technology and it’s capabilities and use cases
When you combine all of these factors and others it’s not really that unusual that it hasn’t happened yet
good thing you’re here! have you sent them your resume yet?
So I watched this video because I didn’t hear that a human beat the AI, but this video is honestly kind of misinformation.
The team developed an adversarial model against the presently best model for go and then had a human play the strategy it played. (It’s honestly kind of scummy behavior).
The best Go model is trained using deep reinforcement learning not with adversarial networks so the example in the video isn’t really genuine.
By that I mean, the model they trained is specifically designed to counter the best go model only, not to be a well rounded go playing model, whereas the best go model is designed to be well rounded at playing Go.
The model in the video is a 1 trick pony which can’t do anything besides troll the best model, whereas the best model is actually good at all go play styles and can beat a variety of oponents.
The best part is you can simply retrain the best model to incorporate the adversarial network to create an even better model that won’t lose to that gimmick.
Also none of this would apply to wow which is a totally different use case and style of ML engineering than for Go. Primarily, Go is a graph search model finding the best move whereas balancing in WoW is a much simpler mathematical regression model.
(to put that in the context of wow, the Go algorithm would be comparable to a model which predicts the best next spell in your character’s rotation, whereas, the balancing model I am describing is just recalculating the spell power coefficients on all the spells given the log data for player performances so that they are more even).
Think they’re getting better at this, it’s just after the fact and some wildly out balance stuff gets pushed live.
Probably because it wouldn’t really help in accomplishing what their goals are with how they design classes. Sure, an AI model could be used to run a number of iterations of tests to seek out how a new spell could affect the performance of a class, or find something that could potentially be theoretically game-breaking, but so far, anytime something has happened in the live game that seemed like a ‘balancing issue’ (like with the multiple augvokers + 2 UH DKs bit that happened last week), it’s been cheaper to make the tweaks to class balancing as the imbalances reveal themselves.
It’s more important to design and tweak the game based on how human players are consuming, playing, and responding to the game, not with how an algorithmic model showed how players could potentially do x and y things that deviated from a desired norm which could potentially threaten the balance of the game.
I’m with what Piddy said in their posts. From a corporate standpoint, it’s cheaper for them to let players run the sims and come back with the feedback on what they want to see from the classes, and it’s likely more profitable to design classes not around refined balance between class and specs but rather keeping them balanced enough to player tastes at the time.
That’s not even beginning to cover the issues developers are having with AI being pushed into the mix of their industry and how it affects their work and their jobs. That’s an entirely different can of worms, but not something that should be overlooked.
1 Like
I understand your points completely, which is why I’m trying to approach it form a positive standpoint.
Having a tool which can just balance the classes with a click of the button would allow devs to spend less time worrying about the interactions of spells and more time developing fun abilities and mechanics because they can use that tool to balance those new abilities very quickly.
This results in a faster development cycle for new content which leads to more content which leads to more MAUs.
So so so many people are watching GD forum without active account waiting for good news about class balance / interesting new spells and talents.
I resubbed only because people were complaining about healing being hard and I wanted to get in on that.
If blizzard actually spent time to balance things properly and then open themselves up to be free to work on cool abilities and stuff then so many more people would come back to play.
Especially with pvp.
Wow pvp is fun (when it’s more balanced and there is diversity in the meta).
So many people would come back if blizzard could get a handle on balance and then introduce new fun toys
But counter point. Teams have finite resources and budgets.
As you stated.
How many devs would they have to get rid of to pay for this?
If so many are gotten rid of…are they going to have the free time to work on all the cool abilities, etc?
Bro they are so desperate they are giving away free boosts.
If the game was well balanced and they had time to fix talent trees instead of worrying about mages with the new augvoker then players would be BUYING BOOSTS to come back and play a FUN GAME.
And honestly, there are so many people trying to get into ML engineering that there is a ton of cheap talent they could pick up as just another member of the team.
The value is there 100%
What misinformation is there? An amateur beat the AI. It shows that these models are developed in a way where the AI literally doesn’t understand what it’s doing. In effect, it’s just “this is probably the best move because that is what my experience tells me.”
If an AI learns poorly, then it reinforce poor behaviors. Unexpected things happen, that’s fine. What AI-based balancing will never achieve is a “desired rotation” – if the gameplay isn’t fun, the balance doesn’t matter. It’s effectively regression on 20 years of WoW theorycrafting.
2 Likes
Irrelevant to anything I said?
Or maybe because they had to get rid of so many devs, they wouldnt have time to do that.
Dang. We went from it being really expensive to really cheap now.
In just a few forum posts.
Amazing.
1 Like
Sorry I’m not going to argue this point any further. You can research the terms in my precious response and try to learn about why I said what I said.
Especially because you are parroting stuff from the video without any actual understanding of Ai.
Of course AI has no understanding, it’s simply just applied statistics my man.
It’s a tool just like anything else, a very very potent tool. But just like a drill that can run out of battery or break the drill bit, it has flaws. That doesn’t make it any less useful or life changing