cross-posted from: https://piefed.ca/c/[email protected]/p/1004331/ai-driving-game-players-are-burning-through-so-many-tokens-its-developer-had-to-take-out-a
The developer behind Teach My Little Sister How to Drive, a driving instruction sim where you instruct a generative-AI-powered, dynamic NPC, has been spending $1,000 dollars a day to keep the demo working.
Teach My Little Sister How to Drive relies on Google Gemini and ChatGPT to power its dynamic NPC, which follows your instructions, reacts, and disobeys, but the developers are looking into ways of allowing local AI models to step in for players “with sufficiently capable PCs.”
They’re using two cloud-based models for a task that is usually handled by if-then-else statements? I think I know who programmed this NPC…
Why use an if statement that compiles down to a conditional branch instruction and takes a nanosecond or two to execute, when you can outsource the decision making to a paid cloud service with expensive hardware that multiplies matrices with billions of elements to get your answer?
This past week at work we had a server that kept getting overloaded because the customer started heavily using a feature that directly runs multiple files from the command line for each interaction. The feature was built long before I was there, and not thought out very well. Each files loads the same dependencies every time it’s run which takes ~5 seconds under heavily load x 50 scripts trying to run at the same time.
I pointed this out for the 3rd time that day while my coworker who has seniority was trying to figure out thy the server was stalling and kept just deploying updates that chatgpt suggested. Suggested to turn those scripts into reusable modules so we load the dependencies once and fork them to run in the background, but the coworkers robot said forking would actually be more resource intensive than running the scripts, which isn’t true if you think about it for half a second.
In the end, he said “Astra says we should upgrade the instance to <some aws one with crazy resources> and avoid updated the legacy script since they’re working as is”. My boss finally chimed in when he realized the coworker upgraded the instance from one that cost ~$200/m to one that would have cost ~$2,000 with like 60 cores and a gpu for some reason. His excuse was the robot suggested it and he stood by that decision.
Thanks for listening to me rant.
YandereDev has apparently moved on to new projects
The game is also popular in Asia, where there isn’t such a stigma around generative AI, which people in the West tend to hate because it’s a theft engine that sucks up the planet’s energy like Shinra.
Fuck that’s a good analogy
I’m gonna start referring to all of the AI companies as Shinra
Thanks… Now i have to know all about Shinra. Imma get my shovel and head torch
It’s from Final Fantasy. They exploit the planet’s lifestream for power and convert that power into tyrannical control.
Disturbingly close to our own reality.
When will people learn that cloud (aka someone else’s computer you rent) is not a good deal for this sort of task? A person smart enough to write a game should really have the fore-thought to estimate the costs up-front of using an external service as part of it.
On the other hand, I’m happy if people token-max the FREE tiers of “AI”, to drive these a-hole companies out of business faster. It’s just unsustainable and the big cloud companies are burning money way faster than they can realistically charge for it, so the more free ‘burners’ going, the better, in a twisted way. Bankrupt those MFers, no mercy.
A person smart enough to write a game
It’s really not hard to write a game these days. (Good or high end games are a different story). These pictures look 100% asset flipped using a driving sim tutorial project. There is zero smarts required. I’d be surprised if they didn’t just download a demo project and plug in the LLM.
when will people learn
Never
Sounds like a perfect microcosm of the whole stupid thing. It’s basically the “SliceLine” business model, from Silicon Valley. (Where they were buying Dominos’ pizzas and putting them in their own boxes and reselling them for less than a Domino’s pizza.) You can’t make a profit, can’t even not lose money with every customer. But you can burn venture capital to offer your product at a loss and then use your increasing revenue to attract more investor money, while promising a new version of the service that actually will be profitable, right around the corner, real soon now- and hope to get bought out before your bullshit all falls apart.
That sounds remarkably close to what AI companies are doing already, now you mention it.
NPCs in games may be one of the very few actual good uses for AI. But not using the cloud obviously.
Theoretically.
In practice, 95% of consumer GPUs are too busy and too small to run an LLM good enough for dialogue, or a decision model smart enough to be warranted over hardcoded logic.
CPUs are too slow. NPUs aren’t universal and are, unfortunately, all but useless with modern software frameworks.
I am very much for LLM NPCs; there are some fantastic mods for such things.
But in shipping games? That can’t be jank?
It is not time. And I would never buy a game with a hard cloud dependency for SP. We will have to get past the RAMpocalypse, at the very least.
The graphics processor is busy doing graphics processing while playing a game. All supporting features of game development: logic, physics, npcs, only run on cpus.
NPCs in games may be one of the very few actual good uses for AI
No. No, they’re not.
Hey, you. You’re finally awake. You were trying to cross the border, right? Walked right into that Imperial ambush, same as us, and that thief over there.
Yes, exactly. You know, I really wanted to cross that border to reach my family. My mother is very sick and needs my help. I brought some things along to help her, too… Some Trama Root and a Thistle Branch. The only thing left I need for her is a python script that reads a six column csv file that contains the start and end dates of our business quarters, the locations, the spending and income of those branches and their name and builds a nice graph to visualise the development each of those branches went through over the last ten years.
I thought we all learned this with the Batman Arkham games. “Real” AI would have made those games unplayable Souls-like nightmares as the enemies learn every single trick or something. We only got the dark knight power fantasy because of the very basic state machines they used. Like I teach them being praised at the time.
I think that it would be highly dependant on how the AI model is trained and what genre of game it is implemented into. For example, games that more heavily rely on dialogue or social interaction.
Take Danganronpa, which is a series if linear narrative games about students trapped in a school forced to kill each other (and get way with it) to escape. You as the player have to investigate each murder and arrive at who the culprit was to win the game. They are fun on your first playthrough, but on any subsequent playthrough you already know the solutions. Like playing Clue but the game only ever has one outcome: Mr. Green, in the Library, with the Lead Pipe. You wouldn’t even have to open the game box and you already know. Absolutely zero replayability.
With an AI in such a game, you could have a new gamemode, like a New Game+, which players can play after the hand-crafted story. This is where I think something like a very carefully trained LLM could be useful, because it could create these kinds of situations at random, making them unpredictable and therefore adding replay value. Now, I don’t think AI is currently sufficiently advanced for this use, but if it continues to develop I could see that niche making good use of it.
Nah
First of all, video games, including detective adventure games with randomised riddles to solve and different endings, have been done for quite some time, long before llms. See e.g. https://en.wikipedia.org/wiki/Blade_Runner_(1997_video_game) . You don’t need a language model for that. Just be passionate about the project.
Second, the whole idea of a video game like that is to tell a story. The last thing I would want is randomised bullshit to click through, wondering all the time if what I’m reading is even still within the boundaries of what the developers wanted or if the llm was already going off the rails. And that’s not even considering the shale feeling of consuming not some created, curated story but merely the condensed diarrhea of stolen work.
No thanks.
A locally run model with quest components and npc dialog as its training data could be tuned in such a way to make radiant quests more unique and engaging. The scenario you’re putting forward assumes a frame around a general purpose LLM, which would be a very bad idea (as the OP article shows.)
