"we have 700m people in our network. We think it's not unreasonable to at least get 50% more people as a result of personalization, culturalization, deeper, more immersive experiences." EA exec on AI
"we have 700m people in our network. We think it's not unreasonable to at least get 50% more people as a result of personalization, culturalization, deeper, more immersive experiences." EA exec on AI
Today Electronic Arts CEO Andrew Wilson provided a look into his ideas for the use of generative AI in the company's development processes.
goddamn, did chatgpt create his answers
“It use to take 6 months, now it takes 6 weeks.”
Amazing, i bet the devs and artists working on it really enjoyed that 4 month vacation they got now /s
21According to Wilson, every time a game is made bigger, deeper, or more immersive, it attracts more people and they're more engaged. He also argued that it's possible to use personalization, culturalization, and deeper, more immersive experiences to draw 50% more people and to make them spend 10 to 20% more on the games thanks to generative AI.
you’re right, I can’t decide if this oblivious fuck of a CEO wrote this trash or if his lazy ass just got gpt to do it. I don’t personally know anyone who still buys EA games, and I don’t think making even bigger budget games with even more scope and even less depth (thanks to generative AI making everything look like trash when you look too close and LLMs turning the story into a nonsensical pile of cliches (though EA does specialize in that last part already)) and the same godawful gameplay as every other EA game will lure any of us away from the low-budget indie games we buy for $10-$20 and play for the next 8-10 years
17Would I buy a FIFA game if all the players had extra fingers, though? Mmmmmmaybe.
15Gordon Banks seed #392858324. Fingers: 9. -2 defence. Reroll Gordon?
3
that whole fucking article reads like it was written by a LLM
14
Here's an idea.
Make better games.
15I wonder if that 30% is an accurate assessment of the time saved when they stop taking the llmcode inputs seriously in review
14If every other measure is equal, and there are 30% more human fingers created, that means EA is 30% more efficient, by this measure. 13-fingered soccer players.
13i don't know where that number could actually have come from. i mean it came from his arse, but which analyst report did he have ready to vaguely justify it if he ever had to
10just vibes
below 30% is "why bother", above 30% is suspiciously "wait how". 30% is nice and cozy, with not too many questions
totes the kind of efficiencies you could squeeze out by checks notes "eliminating human friction on delivery"
imagine all those pesky questions from minions going away! progress!
7
700 million Effective Altruists? That's one big cult!
Oh it's the other EA! You had me there for a minute.
12Developer productivity can definitely increase with codegen tools. The example of a stadium (or player animations) is also feasible assuming they move from developing a stadium rig + code to building something that builds a stadium rig + code. Most of this shit is gluing boilerplate together with minimal thought until you need to do things like optimize, solve new problems, or refactor the codegen spaghetti a few levels deep.
The more you rely on codegen, the more talented your engineers have to be because the easy shit goes away. That means it’s not necessarily cheaper because you shift your problem domain. You can definitely get a boost. 30% seems way too fucking high without serious in-house tool development, which doesn’t sound like their plan. Tabnine trained on EA’s shit code isn’t going to make them that much more efficient.
Also 50% more gamers? Yeah, bullshit. Sports games are going the way of streaming services. Everyone wants to do their own thing. Your audience there can’t go up much unless you improve access to consoles for your games or increase fans that like games of the sport they’re a fan of. We’re 30+ years into that; not really sure there’s much of a market with current tech. On the other end, shooters like CoD are also probably saturated. I just don’t see how the next Star Wars or Sim City flop could bring in 300mil net new users much less, say, a new Fortnite. I find it very hard to believe there are that many more users in the total addressable market that don’t touch EA games that would touch a new one.
11A long time ago amateur/indie game devs looked to procedural generated content as a sort of holy grail to generate games with more interesting content quicker. (it was esp common in the roguelike spaces). Didn't work out that way, stuff wasn't that easy to setup, and often was pretty shallow, buggy, and bland. But sure, this time it will work. (PGC still has its uses btw it just isn't a holy grail).
11I didn’t read that as procgen so much as codegen. With the right team, it’s pretty easy to set up. Tarn Adams has been doing it solo for almost two decades. However, you are very right about how shallow and bland it can be at scale. TES II Daggerfall is one of the best examples of vast reach with procgen that gets real stale after the 100th dungeon that looks damn near the same.
7Pretty easy to do, just takes 2 decades. ;). https://www.markrjohnsongames.com/games/ultima-ratio-regum/ Ultima Ratio Regum is a similar project also taking a decade of work now.
4
Well, all of the most successful roguelikes of the past decade use procgen. Slay the spire procgens the map and encounters. Darkest Dungeon was all procgen and it had sooo much replayability. The Binding of Isaac is all procgen.
6Those are Roguelites [Angry man vs cloud noises], yes people use them, but in limited matters and it isn't some magical fixing thing that can do grand things. The bosses, events etc etc are all still setpieces. There was this time where people expected it to be able to do more, a magical time of hope and wonder, before Radiant Quests. (Several games actually did this btw, warning forever, all the games from Sodak Entertainment. It just has a small lacking thing, even if the games are cool, and anybody interested in this should have played WF and some of the Sodak games), or look at Crawl Stone Soup, how they use random dungeon generation combined with 'vaults' setpieces of terrain which can have special (and even unique, careful around those statues of Hellephants) effects added to them.
But there was a certain type of hype around the method. So much that it even had a wiki created specially for the method. Basically im more talking about the hype around it before people had figured out the limitations and effort involved. A bit like XML, the hype around that was also quite something. Obv XML has its uses, it just wasn't the magical thing people said it would be.
8
wow, taking 6 months to build a static model! adding a thousand variations of a running animation! it's good to know they're so commited to making games fun!
10Hardly a surprise that Android Wilson, himself animated by a host of simulated intelligence procedures and a cluster of servomotors, would try to remove humanity from any and all processes
9I don’t get it, can’t they just build more customers? Or is that the plan?
8"He brought up the example that in the past building a stadium for a sports game took six months. In the past year, it took six weeks, and it's not unnatural to think that in the coming years, it'll take six days."
Does anyone know what the hell he's talking about? Even in Qatar and China, not countries known for their sterling commitment to workers' rights, building a stadium takes 2-3 years.
6Permanently Deleted
14Based on my experience, this is the correct interpretation. Also prior to the required use of Frostbite (fuck that’s almost a decade ago now) you had some complexity trying to get everything working across all your shit and contractors and platforms. After required use of Frostbite everyone had to relearn shit. EA is probably hitting its stride across acquisitions now with that (if not the last several years) so, assuming delays aren’t due them RIFing their talented knowledge holders, some of the efficiency gains are actually from engine familiarity and not codegen.
6