Topic: AI Usage in Video Games

@FishyS Again, I think it's being oversold on both of those things.

The Navier-Stokes solution is controversial because the proof of concept was largely the work of human beings, with the AI used to number crunch the formal version. It's not exactly the breakthrough that was claimed:

https://www.abc.net.au/news/2026-09-10/openai-navier-stokes-m...

Similarly, the idea that LLMs are particularly good at hacking is mostly built around an incident where an AI escaped from a sandbox containing a known vulnerability that it had been trained to exploit, and subsequently attacking external sites containing the same vulnerability.

For anyone involved in software development, it's an everyday occurrence that such vulnerabilities are discovered, and there's typically a fairly short window before a patch is pushed out. We've had malware get half way round the internet during the pre-AI era, so it's not really anything new and they're essentially spinning human incompetence as an AI capability.

Obviously, there are a lot of uses for AI tools in software development and I wouldn't want to deny that. Still, getting thirty bucks a month from every professional coder isn't enough to prop up the megabucks companies, and they desperately need to find a broader user base.

@UpsideDownRowlet I feel like the useful and semi-benign use AI companies are currently advertising is to solve all scientic problems and make new technology and medicines. The number of scientific papers has massively surged in 2026 because of AI analysis assistance, but whether that translates into real life practical improvements remains to be seen.

As a side topic, I do not understand why anyone would want to talk to a chatbot. I just don't get the appeal. Using it for coding or research, sure, it's useful, but just chatting with it seems weird.

@UpsideDownRowlet @VoidofLight fantastic write-ups from the both of y'all. they mostly match my own thoughts on ai and genai, minus the insults and venom lol.

maybe that's why the way nl articles are written bothers me so much now. they're so sterile and bland and lacking in personality that they have retroactively become ai slop style writing in my head. again, i write like this because i've found most llms have a bit of a hate boner for typing anything out vaguely casual. i get to keep my grammar without looking like a bot, win-win.

not saying the people at nl need to write like this, but.. man, you ever think about writing your own gaming articles on your own blog? like just out of spite of such bland writing? you're a gaming news site, i doubt people will sue you if you give opinions anywhere else besides the opinion pieces. let me hear what you really think! give me more stuff like that one video on the 8-player mk8 update that has actual attempts at humor and personality! GIVE ME A REASON TO CARE ABOUT THE NEWS ON A NEWS SITE DAMMIT

@FishyS Again, I think people mostly just pick chatbots for the convenience of it. In real life conversations, there's some level of risk of humiliation, rejection, or criticism. While it's necessary to experience these feelings to grow as people, our brains sure as heck make us not want them. But since our brains didn't evolve to account for AI models that do nothing but shallowly affirm, we may (and I say 'we' just to talk about people in general; I don't expect either of us to do this), particularly at times when we feel vulnerable, find chatbots appealing.

@squiddu-real Thanks a lot! On the note of the sort of staleness of NL news articles, I definitely agree that they're lacking in personality. I never really read news articles in full because of it, always just skimming for the key bits of information I don't already know. I do find your comment that no one will sue if authors give some of their thoughts in news articles a bit funny considering I just saw a comment in the most recent article on the Level-5 AI controversy calling out NL for "bad journalism" for inferring in the article title that the company's reaction to backlash constitutes a panic.

@UpsideDownRowlet clearly that person doesn't know how to have fun :3

thread where i rewrite nl articles with extra spice and that's it. by the way, nl, i don't know if you noticed, but several old articles are tagged but don't actually show up in tags. some 3ds articles can only be found in the ds section, despite still being tagged 3ds!

@Matt_Barber Your points are valid but that doesn't mean the AIs haven't been impressive. The 'destroy society in the next 5 years' theories are based on LLMs continuing to advance quickly, not with current capabilities. I only gave the math example to show how much improvement LLMs have made in some areas very quickly since it wasn't long ago we were making fun of LLMs for failing at very basic math. There have been quite a few AI-assisted math results in the last few weeks and they have all involved human work rather than being purely AI but some of them also clearly sped that work along years faster than it otherwise would have. Same with zero days - they have always existed and humans can find them, but if LLMs can locate and exploit them super fast in the future, that is a new concern; it's not about being better than humans, just faster. And possibly having zero concern about breaking laws.

Whether LLMs will continue to improve at this pace or quickly plateau is an open question. Money is not infinite ( I agree with you it is in an unsustainable bubble right now), so future improvement may come down to whether the AI companies are able to get the LLMs to improve their own efficiency or not.

@FishyS Sure, but perspective is called for. It's not like a University mathematics department is going to get billions of dollars of investment for solving these problems. Rather, it's a million dollar prize and the prestige of having done it.

For the AI companies though, they are treating it as the justification for that and doing their best to erase the rather more cost-effective human component in the process. More generally, when you look at all of the supposed achievements of LLMs, it's worth asking the question of whether billions of dollars of investment elsewhere could have achieved similar, if not superior, results.

Like I said before, I don't think there's no case for it at all. Just using LLM tools for autocompletes and fixing bugs is a pretty easy win across the board. After that though, results van vary, and it's a common enough occurrence to see developers talking about how they'd burn through half their credits for the month on a single request that doesn't really achieve anything.

A lot of "progress" is seemingly being achieved only by throwing more computation at the problem, which someone is ultimately going to have to pay for.