Quoting: Kimyrielle

Quoting: Purple Library GuyBut in reality, as far as I can tell they're actually a technology that has been around for some time, which over many years was developed to the point where someone with seriously deep pockets felt it was worth putting the muscle in to get a really huge data set shoved into one. So rather than a disruptive newborn technology, it's more like a fairly mature technology which has now been brought to the full industrial scale.

Yes, and no. The math has been there for a while, true. But only now we have the serious computing power to actually train these models to the degree they're useful. Can't train GPT-4 on a cluster of 386s, I guess. ;)

They got what they got currently mainly by getting a bigger (data) hammer, and there aren't a lot of bigger hammers left to get, so I'm not convinced they'll go that much further.

I don't know about that? Looking at NVidia's AI chips, every new generation seems to be ridiculously more powerful than the last. I guess we will throw bigger hammers at the problem for a while longer, and we're already at a point where eliminating major issues with the tech is within reach. We will never be able to 100% eliminate hallucination, because that's not how the math works, but reducing it enough is sufficient. Humans hallucinate, too. The AI only needs to be as good as them, not perfect. We should be there soon enough.

I'm not saying they're bad technology. Really, they'd be pretty cool if weren't for all the relentless grift and hype and the likelihood that anything useful they do will be used by really rich people to make everyone else's lives worse. I just think they're a lot closer to a plateau in terms of their capabilities than a lot of people expect.

I keep reading papers on improvements and new approaches for existing tech on a daily basis. We're not even close to plateauing. And while I share your concern about AI being monopolized by Big Evil Tech, so far it hasn't happened. There are a lot of open source models around that can compete fairly ok with the closed-source cloud services.

Just to be clear, when I said bigger (data) hammer, I meant the amount of data being used to train the model, not processing power, which doesn't seem to be the limitation. The top ones currently have basically scraped the whole internet, so I'm not sure there's a whole lot more to scrape. If anything, going forward there may be less, as copyright holders block such uses and privacy models pay attention to the issue; as well as worse, as AI models increasingly scrape AI content.

When I say plateau, I don't mean to say nothing whatsoever will happen; plateaus aren't entirely flat. But there's a difference between the progress in aviation from the Wright Brothers to the 747, and the progress in aviation since then (not to mention regression; 737MAX anyone?). I'm saying that "AI" bursting on the scene all of a sudden makes us feel like this is the Wright Brothers time for this kind of technology, when it may instead be the 747 moment and the earlier part was just quiet. There will probably in the future be other kinds of "AI" based on different ideas that give results that are intelligence-like in more important ways, but I think it's likely they won't be souped-up ChatGPT.

View this comment - View article - View full comments

Quoting: KimyrielleThey will not do that for too much longer.

This is the assumption I'm not sure about. We tend to think of these AI things as a disruptive new technology just at the beginning of a climb to amazingness. But in reality, as far as I can tell they're actually a technology that has been around for some time, which over many years was developed to the point where someone with seriously deep pockets felt it was worth putting the muscle in to get a really huge data set shoved into one. So rather than a disruptive newborn technology, it's more like a fairly mature technology which has now been brought to the full industrial scale.

So now we have them with the really huge data sets, and I think it's shown us some impressive things about this kind of "AI" technology, but also some limitations. They got what they got currently mainly by getting a bigger (data) hammer, and there aren't a lot of bigger hammers left to get, so I'm not convinced they'll go that much further. It's like the way self-driving cars looked so promising and then their capabilities kind of plateaued and now they don't get talked about that much and one of the pioneers of the field is now working on simplified self-driving for mining and other industrial site vehicles because he's decided the full consumer self-driving is not a problem solvable with the kind of technology current self-driving is based on.

I think they're also unusually difficult to improve. Normal technologies have known characteristics, which you can then tweak and build on. These "AI" technologies, including the self-driving cars, are about a kind of forced evolution process which results in black boxes. You can't really tweak and build on them because you don't understand what they do in the first place.

I'm not saying they're bad technology. Really, they'd be pretty cool if weren't for all the relentless grift and hype and the likelihood that anything useful they do will be used by really rich people to make everyone else's lives worse. I just think they're a lot closer to a plateau in terms of their capabilities than a lot of people expect.

View this comment - View article - View full comments

Quoting: pb

Quoting: Salvatos

If in a game scenario you bargain, AI can remember this the next time.

Holy shit, they just invented persistent game states! Now I understand why everyone is hype about AI. Sign me up!

Imagine this: you buy an RPG game and the NPCs view and treat you depending on whether you bought it for a full price or on sale. If you bought it at -90%, your reputation is minimal and everyone sees you as a pariah. On the other hand, if your reputation is balanced, every cosmetic DLC you buy makes you more of a celebrity in the game's world.

An interesting/horrible idea. But it still wouldn't require AI in any way shape or form.

View this comment - View article - View full comments

This looks like the kind of thing I would enjoy in theory, but in practice I'd get stuck on all the puzzles.

View this comment - View article - View full comments

Quoting: JarmerWhoa, this is a blast from the past! I think I've had this on my wishlist for ... actually I can't even remember, 4 years? More? I used to have this high on my anticipation radar, since the demo was great and the entire setting was just fantastic, but I guess it somehow got stuck in dev hell if it's still EA after all this time.

I will still keep it on my wishlist, and I hope they can get closer to a release.

Is it in EA? Huh, I guess it is. Weird that they're adding a named content update while they're still in EA.

View this comment - View article - View full comments

Didn't they used to call this "jumping the shark"?

View this comment - View article - View full comments

Quoting: alexleducIn other news, the stat counter worldwide desktop market share survey has been plummeting back down from its 4.05% milestone in March. It's at 3.77% in the May results.

Drat.

View this comment - View article - View full comments

Quoting: Linux_Rocks

Quoting: PenglingI just gave it a try in my browser, and this is lovely. :grin: It reminds me a little bit of running Dogz in windowed mode, back in the 1990s!

We had Catz and Dogz 4 on PC back in the day. I made the most mutt animals I could by breeding all the breeds together in a giant family tree (and no, you couldn't make a Great Dane mate with a Chihuahua. It just wouldn't let you do it. lol)

I'm sure IRL a boy chihuahua would be willing to try. Tiny dogs are all mountain climbers.

View this comment - View article - View full comments

I dunno about the Dragonball Z game--I heard the boss fights take hours. :tongue:

View this comment - View article - View full comments

Quoting: Liam DaweArticle updated, Paradox Tectonic is being shut down.

Whoa.

View this comment - View article - View full comments