It would also require more optimized builds, leaving little wiggle-room for mistakes.

How?

For any set of stats, it wouldn’t matter if you got there at level 2 or level 20. Mudcrabs built for an optimized level 10 wouldn’t spawn just because you are level 10. They’d spawn when you’ve spent as many points as you would have, had you optimized to level 10.

The real issue would be “the Draugr are training” kinds of builds. If you put half your points into personality… the bandits did not.

you simply get a flat 12 points to spend every time you level up.

… that’s the wrong problem entirely! Complicated leveling was just another minigame to optimize.

If they wanted it to be painless, they could’ve leveled the world according to your total stats. Attack the player based on how much their numbers went up, instead of how often they went up.

The deeper problem in Oblivion was that you could blast goblins with a single fireball at level one, and then when you got stronger and put all your points in kill-stuff-harder attributes, the same fireball against the same goblins just tickled.

Wait what? They even said they fixed the leveling, in the announcement video! This was the one thing everyone agreed was broken, and it barely mattered which mod you chose to fix it, because all of them were an improvement. How’d they fuck this up twice?

Talked shit to some dunmer, didn’t know he was from Maar Gan.

“We have those?”

Never give Nintendo money.

That’s what I’m on about. We have the technology to avoid going ‘hold up, I gotta get something.’ There’s supposed to be a shitty version that’s always there, in case you have to render it by surprise, and say ‘better luck next frame.’ The most important part is to put roughly the right colors onscreen and move on.

id Software did this on Xbox 360… loading from a DVD drive. Framerate impact: nil.

They had trouble increasing memory even before this AI nonsense. Now they have a perverse incentive to keep it low on affordable cards, to avoid undercutting their own industrial-grade products.

Which only matters thanks to anticompetitive practices leveraging CUDA’s monopoly. Refusing to give up the fat margins on professional equipment is what killed DEC. They successfully miniaturized their PDP mainframes, while personal computers became serious business, but they refused to let those run existing software. They crippled their own product and the market destroyed them. That can’t happen, here, because ATI is not allowed to participate in the inflated market of… linear algebra.

The flipside is: why the hell doesn’t any game work on eight gigabytes of VRAM? Devs. What are you doing? Does Epic not know how a texture atlas works?

Guinan: “More?”

Data: “Please.”

The episode, summarized:

Nobody listens to Worf.

Nobody listens to Worf.

Nobody listens to Worf.

Nobody listens to Worf.

Nobody listens to Worf.

Nobody listens to Worf.

Data listens to a deck of cards.

I think they mean whatever’s handling the model. A program into which you feed this inherently restricted format, so it takes advantage of those limitations, in order to run more efficiently.

Like if every number’s magnitude is 1 or 0, you don’t need to do floating-point multiplication.

Oh sure, Google only fucks with how texting works, whether my car GPS cooperates, what the entire UI looks like…

I know this specific abuse was not Google’s fault specifically. But it’s the same god damn thing. Nobody asked me if I wanted that. They just shoved it onto the machine I paid to own and pay to use, like it’s not mine enough to deserve respect, let alone control.

It’s Metro’s fault.

I fucking hate that three separate skeezebag companies have free reign to insert software on my extremely personal wireless pocket computer.

The latest Google update will make your phone more secure if you don’t touch it…

Fuck entirely off. The last time my phone surprised me with an update, it was ads on my goddamn lock screen.

Thanks, I stole it.

It’s trinary, and I understand why they instead say “1-bit,” but it still bugs me that they call it “1-bit.”

I’d love to see how low they can push this and still get spooky results. Something with ten million parameters could fit on a Macintosh Classic II - and if it ran at any speed worth calling interactive, it’d undercut a lot of loud complaints about energy use. Training takes a zillion watts. Using the model is like running a video game.

All the cool kids left.

… and reporting itself can be a bannable offense.

Thank you. This shit is cool, no matter which grifters insist it’s the next internet or whatever.

We have proven that data and compute alone can produce deeply spooky results. LLMs should nnnot be capable of answering questions. They’re glorified Markov chains. And yet: asking “what word comes next?” is correct-ish, often enough, that we have to talk about the limits of their knowledge and reasoning. Diffusion models genuinely remove the parts of the marble that don’t look like a statue. Combined, you can just type “Han Solo scene but Jabba is Shrek” and it’ll do a half-ass job of rendering video of that ridiculous premise. It’ll do a lot more if you provide a half-ass job and tell it to remove the parts that don’t look right. It’s straight-up science fiction and people are performatively incensed that it’s William Gibson instead of Isaac Asimov.

Their use of diffusion to mean wider distribution is an unfortunate choice.

… this whole article still treats the technology as slightly magic, huh. Disappointing.

Look, the future is in smarter questions for what a neural network should do. We’ve solved training in a way that a shitload of effort will definitely produce results. We don’t seem to give a shit about licensing data, which is fine by me, because eternal copyright blows, and training is transformative use. So fundamental shifts in how well a pile of linear algebra does what we want will depend on knowing what we want.

Even just using the two formats anyone cares about, you can use diffusion on text, and it’ll gradually unfuck whatever’s written. That’s a significant improvement over guessing the next word and proceeding like it’s carved in stone.

… the foreign-hosted servers, or the models themselves?

There are lingering questions, however, around whether DeepSeek engaged in IP theft to create some of its more competitive models.

They did, just like everybody else.

OpenAI has alleged that the Chinese lab distilled its models, violating OpenAI’s terms of use.

Oh that’s fucking rich.