It is certainly explainable.

OpenAI was given half a trillion dollars to 'develop' AI with. It's called project Stargate.

The first thing they did with it was wave the check around and promise to buy 40% of all wafers produced globally. Wafers are the precursors to memory chips. OpenAI doesn't need wafers; it needs working memory (either ram, Vram or SSD). They don't manufacture anything so they don't do anything with the wafers. They just don't want anyone else to have them because the competition can use the memory.

But capitalism does what it does, which is to chase profit, and ever wafer manufacturer was happy to get a piece of that half trillion dollars. It's no surprise that the first to abandon memory production were Micron (the company behind Crucial, which is the finished product division for consumers) and Sony, also producers of wafers.

There was never any guarantee written down anywhere that gpu prices would remain stable. That would actually be going against the laws of capitalism. If it hadn't been AI, it was going to be something else upsetting this balance.

I was part of a project that tried. If you just want to get rid of dialog trees local processing might be an option, but for an actual reactive world your looking at possibly hundreds of dollars a month in rented compute. The deeper problem is the current generation of LLMs are just too unstable over time, and frankly, too dumb to maintain its persona and keep up with a dynamic fictional world. Each hallucination amplifies the disconnect from the fictional reality, and trying to reset it changes the presentation of the persona. And if the humans are actively trying to mess with the AIs or if the AIs are supposed to consider something without blurring it out, its all going to crash and burn.

Of course, its a little concerning that these agents are too unreliable to be a character is a fictional video game, but apparently reliable enough to be given access and authority in the real world.

I agree, and it's very possible. I think the real limiter is that consoles don't have the RAM necessary to load the models to enable it. Maybe in the future, when models become more efficient in memory usage, or more likely, the RAM shortage is resolved.

I dont know how realistic my dreams are there.

Yeah, as I said in another comment in this post, I don't think LLMs can do anything like that, and I think what you're describing is going to require something that doesn't exist yet and that we have no reason to think will exist any time soon. (And I'm super pissed that Nvidia et al are raking in billions in dirty grifter money on the contrary promise -- which it can't keep, mind you.)

Every coworker I've seen that uses AI code tools heavily is bad. They produce (or at least push) nonsense code they don't understand.

I would rather have a team that goes slowly and understands what they're building than a team of excitable slop pushers going a thousand lines a second.

I'm a senior engineer of 10+ years. I do not, nor will I ever, use AI to write code. Fuck that shit. It makes everything worse in countless ways. So much garbage I have to deal with at work now because of it, and it absolutely has made no difference in velocity. I take that back actually, it's slowed things down because now there's a lot more "what the fuck is this shit" reviews I have to do now, which just slows everything down. Like just the other day where someone on my team submitted a MR where the AI wrote a pre-commit hook that ran a script to parse the source code of our entire monorepo to scan for occurrences of a few functions to make sure they weren't used.

I have to deal with so many "ideas" now that no sane person would ever come up with, because they are fucking stupid.

Oh, and I can't tell you how many times I've had to tell people to remove absolutely useless tests from MRs. It's almost a joke at this point.

The good news it's already starting to get way too expensive to keep on like this and management is finally starting to feel the pain, so I doubt it really has much legs left. It always comes down to money, in the end, and this shit isn't cheap and will only continue to skyrocket in price as the wildly unprofitable AI companies run out of money and need to start showing profitability.

As far as I know, the current iteration of Steam's AI disclosure is for asset generation in games, not using AI for coding. So, most of your point is kinda moot in the context of the post.

I don't think we can be friends.

Smaller LLM models could be great for expanding dialog options in game.

I don't think LLMs (smaller or otherwise) can ever clear the uncanny valley. The level of adaptability you're taking about but without hallucinations and random batshit-crazy behavior requires something humans haven't invented yet and that we have no specific reason will ever happen. (In fact, I think if we can be said to be on a path toward building the first AGI, I believe LLMs and "generative AI" and this whole hype bubble will be looked back on as having been a diversion/destraction from that path that delayed the advent of AGI.)

Unless/until that someday happens, I'd much rather play, and feel like I'd find much more immersive, a game that generated, say, fetch-quests with something like My {rand("health","Lord","wife","none-of-yor-business",...)} {rand("demands","begs","commissions",...)} you to bring a {rand("tufted titmouse","Gauntlet of Light","turkey dinner","giant's toenail",...)} that you can find in/at {rand("Illsword Manor","The Cavern of Lies","the afterlife",...)} to {rand("the illfated dragon","Fenworth Blurd","The Rusty Scabbard Inn",...)} in {rand("Feyspring Vale","Weston","the northern wilderness",...)}. than something with the problems inherent to LLMs.

I suppose I can agree that I'd be interested in new technologies that make gaming more immersive, but "smart" NPCs that can improvise and riff while not ruining the immersion like LLMs would seem so speculative and far off that I might as well wish to stumble into the possession of a genie lamp a la Disney's Aladdin while I'm at it. Meanwhile, grifters are making shit-tons of money at your expense promising you just that.

AGI has been only ten years away for the last fifty years. Anyone who says it's less than ten years away now is trying to sell you a bridge.

As for the "filling out assets" and "increased diversity at least at the high end", I think you're overestimating the capabilities of (at least current) generative AI. And if you're not, there are still a lot of legal issues with generative AI, and no matter which way those legal issues end up being decided, we can expect shitty results for consumers and competitors. (I fear, for instance, that the output of generative AI may be subject to copyright when big companies want it to be and not when lawsuits go in the other direction.)

In what specific ways? There are 20 comments in the thread you commented on explaining in detail why they think the opposite, but you said "nuh uh AI cool actually" so you must be right.