#Regarding AI and its behaviors
1 messages · Page 1 of 1 (latest)
I totally agree that Solarint is a guru with AI and an absolute chad. However, I believe the core limitation for AI is with UE itself and the large scale of the game. It's a totally different thing than the two PvE/single player experiences SAIN has worked with.
Isn't he currently under contract though? If he is, good luck getting him
Absolutely kidnap, every move matters, no compromise
I'm pretty sure that is the case however it doesn't hurt to ask right? (or kidnap)
It's a totally different kind of ai when you have hundreds of ai at once and 48 players running on a server than tarkovs smaller instances with much less ai and players or incursions local sessions with few players and maybe tens of ai.
That may be, but the other game can have a metric ton of bots on the map and it shows he can scale up. So why not let him try?
AI are server side, so its a completely different system than running it locally
So you don't want an expert to take a look and see what they can do... even if it's not identical to what's been done before?
Y'all arent really writing him off before even giving a chance is wild.
Id rather he tried and fail than never attempt it at all
Its amazing how you extrapolated all that out of my comment
The reply was in the context of all previous replies as well, not just your comment.
What did you intend to mean if not agreeing with Albert pizza?
He never mentioned Solarint, never mentioned anything about giving him a chance at the AI or anything like what you commented, he was just explaining that the AI in GZW is on a far larger scale compared to Tarkov and instances being specific to that game you're in and that Incursions AI is completely locally run compared to GZW's server side AI
More than happy for AI experts to drop into MFG and help them out, but so far we may be completely bound by how UE runs with its overhead and how the AI is server based. its going to require more than just upgrading to UE5.7 but a lot more optimisations so that more complex code can be run and to compensate for lag and number of players
That all may be true (it is, nothing i see here is incorrect) but the fundamental code behind AI behaviors doesn't matter if it's server side or client side so long as both ends communicate updates in a timely way.
Server side AI is probably easier to figure out too because generally servers can run all those calculations much better than a home PC can. All the client has to do is display the result.
Incursion runs on UE5 so maybe Solarint has had a good chance to get some experience in it that he could help but as far as im aware he is fully employed by those devs so most likely won't be pulled away any time soon
The problem comes down to how much the Devs are willing to pay for what kind of service they get from servers, most servers and broken up to smaller virtualised instances that support the games that players are in, but also these servers have to account for all AI in the game and also run their code consistently 24/7 as they are persistent worlds, while also making sure these are reacting and working with 47 other players other than yourself while compensating for multiple different latencies that players are on, remember render distances are out to 400 meters so at any one time you have 48 players who can "see" what AI are doing in a 400 meter radius around them and the AI have to be reactive to that
Its a tough nut to crack. MFG already has that code more or less in the game already.
All that's missing is what the AI does when triggered by a player and that's where someone like Solarint would be perfect.
Yup pretty much, they just need to expand on it, optimise it as much as possible so they can get more reactive and more complex AI, with time hopefully things will improve, one of the things they mentioned is upgrading to UE5.7 will help get them there as they mentioned in the latest devstream, though i swear there is some major bug with T3 AI as i find T2 AI way more deadly
Trying to rack my brains if i've ever really seen a "persistent" online open world game with good AI lol
Blue lagoon has my dogtags laying around more than all 3 t3 pois combined... i agree with your feeling that the t3 duders feels off at the moment as compared to t2 duders.
It actually looks like they are looking for more AI programmers
You're not taking into consideration that the fastest servers are probably rent for ai/llm solutions and will never be rent for games for reasonable price.
Job offers for ai developers at mfg are open for months if not years already
Problem is probably moving since its on-site position
Different kinds of servers. Game servers are way more focused on latency, ai servers are built to process inference quickly not just latency. Different hardware. Pretty sure the best game servers are still their own market not cannibalized by AI yet.
I read Some online game is going offline cuz servers for ai are more profitable. Can't remember the name but well.
Oh wow. News to me. That's pretty wild, I would think they would need to rebuild the racks completely? Must be crazy good money to rent ai compute
The servers for AI and the servers for games are optimized for different things. AI uses GPUs while games use CPUs. Unless they're gobbling up the CPUs now too.
Honestly i wouldn't put data centers reconfiguring to AI passed them
The shoe brand Allbirds or whatever has pivoted from shoes to being AI datacenters, and their tumbling stock market value skyrocketed on just the announcement
Agentic AI is surging and that leans on CPUs a lot
This is what I was getting at, the actual hardware of the rack is different. AI is GPU/TPU/ASIC games are CPU.
Yep, just checked some CPU prices for servers and it's at least double
It may not be for strict ai computing, but something next to it which still provides more money from pumped ai companies than game dev.
Might leave a really nice gap in the market for game-specific servers 💰
Morgan Stanley forecasts that the rise of agentic AI will lead to increased spending on central processing units (CPUs) and broaden investments beyond traditional graphics processing units (GPUs), hinting at a transformative shift in the data center market with potential beneficiaries like Nvidia, AMD, and Intel.
MOE models like deepseek can run on CPUs relatively decently so yeah there's that too. RIP games
First it was Nvidia, now it's everyone else ;(
"Gpu prices are wild, let's pump CPU prices as well"
So both CPU and GPU's are gonna be in such high demand, that's why RAM is massively overpriced for consumer as both CPU and GPU need that
Interesting. I wonder if it gets bad enough games will switch to some kind of distributed compute instead of a server. Latency is such an issue with that tho
Given enough time supply will catch up I tho suppose
Cheating will be such a big issue that it will just be better to go back to offline games
It would have to be peer validated, like a blockchain
But yeah, very data heavy and slow compared to a server. Tricky
That may work. Supply is a bit sketchy as there is an incentive not to increase the production by much as they fear demand may vanish and they can just make the prices high. Foreign markets and startups may help, but silicon is a hard industry to break into.
Yeah. Also a bit of a precarious industry because everyones eggs are in TSMC's basket
YMCT and CMXT both Chinese manufacturers of RAM have 1 fab coming online in the next few months with 2 more by the end of the year, but Samsung/Micro/SK is gonna take them a few years at best to have anything online
AMD are even started to diversify some of their products to Global Foundaries lol
RAM and NAND storage may go up though as they don't require as high precision node manufacturing as dies since they don't scale down well.
Yup they are realistically only on 10nm at the moment
The biggest thing that will pop the bubble will be power generation or water, it won't be anything to do with CPU/GPU/RAM etc
Well Germany is firing up a fusion reactor. That may help
Don't say that, i want the bubble to pop 😄
Wait what?
fusion?
We can get back to reasonable pricing
Yep, they're getting closer to making fusion more practical, setting records, look into Wendelstein(And apparently a commercial one is being built)
We are at best still decades away from having a fusion reactor be net energy positive, let alone economical.
Germany most definitely isn't firing up a viable fusion reactor.
It's not even clear that we can build an economically viable fusion reactor with current materials science.
Well, it's comissioned to be built in 2040
Whether that's the case or not, any hope is good ;(
Lets hope the bubble pops before then, i don't fancy paying 10 times the price for RAM or my CPU's or GPUs 😄
But you never know by then we will all just play AI games
No need for hardware
Let AI push the world back into the analog age >:D
We have really gone off topic lol
I mean it's still about AI...
True
I sold 64 GB of my old RAM on facebook for $600 lol
Kept 32 for my PC and still have another 4 sticks of 16 if I need it
I just bought a bunch of RAM 12 years ago when it was dirt cheap
Yep, that's my point. I'm just thinking wouldn't it be nice if that spot was filled by a certain calibur of person with one specific dude