T w e n t y S i d e d
I’m guilty of hypocrisy and what I consider a grave crime. I buy the newest EA sportsball game or MLB: The Show. As I hit that “Confirm Purchase” button I have a wave of regret and shame every time. I’m an informed consumer of an industry that I genuinely care about immensely, but I still keep doing it because I dearly love the sports being “simulated” and want to be able to participate. Even if only digitally.
EA is evil. I’m not a fan of pretty much any of the AAA publishers and dev companies out there, but EA, in my opinion, is by far the greatest evil. EA does a full combination of the biggest evils that any corporate media entity can do. They buy up and shut down smaller companies. They hold IPs hostage and do nothing with them. They underpay and overwork their employees while subjecting them to hostile work environments. And most pertinent to today’s discussion, they regurgitate the same content over and over because they know suckers like me and casual gamers will buy it anyway. The reason why EA is the target of my ire is because of the Disney-level degree of which they do those things.
The Madden franchise has made some efforts here and there but never truly committed. For instance in ’18 and ’19 they tried doing a true story mode with “Longshot.” It wasn’t great. It was your run of the mill sports game mode where you play as one player locked into one position building a career and legacy. All they did to differentiate it from the bog standard is add some cutscenes with meaningless dialogue choices. Even that wasn’t exactly new, it was just presented in a more cinematic way.
Nothing too substantial has been changed with Madden since they first added a career mode in ’06, but one big change I noticed at one point along the way was the introduction of voiced names. If you have a compatible name for your athlete, the announcers would use it in their commentary on the games. It was clunky and sounded like what it was: a sound file being plucked from a database and plugged into existing templates used as commentary. But it was a cool addition that made it feel more immersive and exciting. That feature hasn’t improved. It’s been many, many years and the commentary still sounds stilted, awkward, and, at some times, laggy.
Sports game commentary peaked at NBA Jam in 1993. Ever since, they’ve done things to make it seem more realistic and closer to a simulation of a real broadcast. The problem is that you have a limited size of database to pull from to stitch together commentary. You might hear the same “analysis” 3 or 4 times in the same match, depending on the game. For instance, when I played the career mode of PGA Tour 2K25 I was impressed. They repeated lines from here and there but it felt rich, accurate, and entertaining. Then, I finished a season and looped back to the same courses and it felt like the their lines were basically just a timed loop. The exact same stuff repeated with different interjections depending on how ass I played at that particular time.
I’ve been playing MLB: The Show 26 since it’s featured on PS+ and I swear the database of voice lines could fit on a single paged spreadsheet. I muted it only a few games in. When I play a new Madden it typically takes me a dozen or so games to turn the sound off. The UFC games are the same story. Repeated “insight” on technique and soulless repetition of the obvious. It sucks, because commentary is a huge part of what makes sports enjoyable or unwatchable.
My dad and I watch every Steelers game. The whole family gathers for dinner, then flocks to the TV to get pissed off and be armchair quarterbacks. We love it. Sometimes we will also watch the game before or after the Steelers game. One thing has historically adjusted plans regarding that other game: Cris Collinsworth. His substanceless ramblings piss off my dad and I enough that we either skip the game or watch on mute. He’s that grating to us. His presence is so awful that it takes something that I love, and bond with my dad over, and makes it annoying. Collinsworth makes it unwatchable for us.
On the other hand, a great play-by-play and/or analyst can make explosive plays feel world changing. Men will remember for years sports moments that they had the privilege to be there for. There’s compilations on YouTube of the biggest and best calls during NHL, NBA, NFL, etc. because of the excitement they bring.
I think it’s fair to say that commentary needs an improvement. Which is where I have something somewhat controversial to suggest. I think AI could be used to make commentary in sports games exponentially better. It would have to come with one hell of a protective contract for the voice actors that protect the rights to their voice and their work, but I think it would make a huge difference. Imagine if EA trained an LLM on sports commentary through the years and trained a program to properly react to moments and to properly blend name pronunciations into it. You could have players type in their name, then correct the phonetic pronunciation to have it used to include them in the fun that some of us have had access to for years, but better because it would be more robust.
I’m not a fan of LLMs when applied to professional creative purposes, personally. I think it devalues the works of actual artists while stealing from them to train themselves. It’s icky. I do think, though, that there are proper uses for LLMs as a tool. And since my idea for an LLM would be trained only by voice actors/commentators who agree to be used in such a way as well as archived audio from the history of the NFL, assuming they agreed to it, the only concern is ecological. And there are ways to produce and maintain data centers that would not damage the community or infrastructure around them if they were willing to pay for it. They have the money for it. They have the influence for it. They could do it. They would never. Unfortunately, that would require real work and real investment and EA has proven that isn’t something they’re interested in.
The use of LLMs and AI is a complex and contentious issue. I don’t use LLMs outside of assisting me with photoshop (selecting complex shapes in photos, creating transparent images from images with flat backgrounds, and sharpening images) and even in that context I’d happily go without it if it meant the other problematic elements went away too. Please keep the comments about the viability and/or stupidity of my idea. Otherwise, scream into the void on whichever social media app/site you prefer, like I do. It’s therapeutic.
The Strange Evolution of OpenGL
Sometimes software is engineered. Sometimes it grows organically. And sometimes it's thrown together seemingly at random over two decades.
id Software Coding Style
When the source code for Doom 3 was released, we got a look at some of the style conventions used by the developers. Here I analyze this style and explain what it all means.
Final Fantasy X
A game about the ghost of an underwater football player who travels through time to save the world from a tick that controls kaiju satan. Really.
The Plot-Driven Door
You know how videogames sometimes do that thing where it's preposterously hard to go through a simple door? This one is really bad.
Bethesda NEVER Understood Fallout
Let's count up the ways in which Bethesda has misunderstood and misused the Fallout property.
Usually a simple solution can get you 80~90% of the way there. Jim and Bob in Blood Bowl, for example, are 90% of the way there. If you’re going to train a model on material provided by a voice actor, the amount of material needed (assuming you don’t use an existing model which already has been trained on a large corpus) would be so large you could just have the VA provide enough material to just generate lines algorithmically. The only exception I can think of is with reading out names.
The problem with using fabricating artificial intelligence will be the fact that players will either need a very powerful computer in order to have commentary be generated anywhere near quickly enough to not slow the game down beyond playable levels, or the generation would have to be done on a server, in which case you get into problems with up and down latency, connection saturation, load balancing, capacity provisioning,… in short you’d need a lot of infrastructure all around the world. Not to mention the fact that players would need to play their game online, even if they play single player, which players generally aren’t overly fond of (think of the Stop Killing Games initiative).
Also, agreed: EA is the worst. Their acquisition of Codemasters means there won’t ever be a DiRT Rally 3.0, and it’s the single worst thing that’s happened to sim racing since its inception. It got me to really despise the excess of consolidation in the video games industry (e.g. Microsoft acquiring Obsidian, Zenimax, Activision,…).
I think this falls into the same problem a lot of discussion of AI has: the answer always seems to be “Yeah, it could do that. In theory. One day.”
Put aside the costs to the environment and the costs to infrastructure like building and powering datacenters. Put aside the fact that it works via stealing the work of humans’ creativity to recycle.
You’ve still got the problem that for an AI to make interesting, varied commentary on a sports match (or sports game) would involve it *actually understanding* (on some level): the sport and its history; tactics/strategies of the game; how to keep the commentary interesting; what it’s said before; other things I’m neglecting. Which is something that – as you pointed out – a lot of humans can’t do. The ones good at it make a lot of money!
I don’t think it’s impossible for a future AI to do what you’re saying, it’s just way beyond what I’ve seen/heard about so far.
I had a big comment written out, but it got lost during submission, I’ll try to restate as much as I can.
Don’t really need fabricating artificial intelligence for commentary, a simple solution will get you 90% of the way there (see for example Jim and Bob in Blood Bowl). Training a model on material provided by a voice actor will require so much material you could just use that material with an algorithm to generate commentary.
Players’ PCs may not be good enough to run a FAI model well enough to get the commentary at real time and having to set up servers which generate the commentary will create technical challenges in getting enough data up- and downloaded and will require load balancers and servers all around the world and so on.
EA is terrible garbage and I hate that they acquired Codemasters so there won’t be a DiRT Rally 3.0. Excess consolidation in the video games industry (e.g. Microsoft buying Zenimax and Activision) is a massive problem that will create many problems in the foreseeable future.
I think Ubisoft is worse, because they ruined Heroes of Might and Magic. Agree that EA is terrible though.
It’s probably a little late, but I’ve played a number of sports games — mostly hockey and baseball, though — and have found that the commentary isn’t that bad, and remember finding that the commentary in NHL 16 was good enough that I was tempted to create a bunch of custom teams and set it up to play out and just watch it play because I don’t really watch a lot of hockey anymore (for the same reasons that Angel would watch it; the games are mostly at night but I go to sleep too early to see the end). The only reason I haven’t is that creating 8 – 10 custom teams on the console takes a lot of time and effort that I don’t have.
But it might depend on the sport. For hockey, describing the action about as well as real commentators isn’t that hard, since you’re constantly talking about passes, shots, saves, entries to the zone, etc, etc, which are pretty deterministic, and there’s not that much time for additional commentary as the action moves pretty quickly and there’s not much stoppage time. And for baseball, again the actions are deterministic and there’s not that much time between pitches, so you don’t have to say much other than “That really puts them in trouble” and “They’re making a great comeback”, and you can always base some commentary off stats. That might not be true for football and MMA.
But I would be one of the ones who doesn’t might the commentary we have, and given that lots of people are probably like me the reason to not improve it comes down to it costing a lot of money to do only to end up with a result that most people won’t really notice and won’t really appreciate.
This mostly makes me think of the old SNL skit. That’s the commentary I want to see. The added benefit being that it would be much easier to implement, because it barely matters what’s actually happening in-game.
Agree that commentary is huge, even the extremely noticeably limited lines in Pokemon Stadium do so much to elevate the experience.
I also have no problem at all with creating a tool to smooth the addition of stock names into scripts, but for such a narrow use, the more effective use of LLM tech would be to generate the database of all the modifier techniques to put in the smaller algorithm the tool would actually use. Indeed, I don’t see why modifers couldn’t be done- one of the main original points of such tech I’m sure was to generate better sounding voices for text to speech, but if you’re working with an existing script and a specific person there are even fewer variables to smooth out.
Names also don’t seem that difficult: storage is more than huge enough nowadays that you could have a database of every name in a language, possibly even on the planet, with a sound file for each pronunciation, and the extremely narrow scope of “allow ‘AI’ generated samples of your voice saying the player’s name” seems reasonable. Although really, I dont’ see why it shouldn’t be possible to generate the sounds for the player’s own name entirely at that point, someone could make a “Name Creator” the same way so many games now seem to have the exact same character creator and photomode modules in them.
You don’t need an LLM to generate scripts on the fly though, and something will eventually go mind-jarringly wrong in there. Like tmtvl said, it should be entirely possible to write enough lines that the game can stitch them together, especially if you allow some voice modulation and have a tool to help the lines flow from one to the next, the problem is writing the algorithm for the game to stitch them together. Games with with well-regarded procedural generation probably already know he right sorts of ratios for how many you need of X common pieces and Y rarer pieces and how often you can repeat one or the other, though speech is different so those might not apply. And of course you could come up with a couple dozen one in a million plays to record specific lines for, which across millions of players someone will eventually run into and make your game look amazing.
Since LLMs are/can be actually good at analyzing databases and generating code, and have already figured out a lot (though not all) of the voice modulation stuff, that’s what you’d have it actually doing. It might be possible to get said machine to write that part of the code/databases, since it already “knows”. Or it might not, since it doesn’t actually know or have self-reflection, or it might take a lot of “training” as you essentially teach your sub-model how to self-reflect its unknown sound generation into usable code.
But overall yes, this does seem like a reasonable intermediate use of the technology to me. Even the most advanced eventually idea will not be as good as the best human writers (though I’ll allow that they might one day match beginners), because even if they have the ability to inject new concepts into the body of existing text they’re working from, they do not live a human experience, and indeed having ingested far more material than any one human ever could, in some ways cannot have a narrow enough experience.
But if the game publishers would actually pay for real writers to write really sufficient amounts of text and then get out of their way, and programmers to figure out how to procedurally utilize it, they could in fact do the thing. But that would require investing money in such ephemeral things as “story”, which most of the people in the business of owning things don’t seem to be able to comprehend or connect with, at least not compared to “oh shiny graphics number go up!”