Live data from Hacker News

WorldClaw Agentic 3D open-world generation at scale

tencent-hunyuan.github.io

81–90 of 101 posts

Re: WorldClaw Agentic 3D open-world generation at scale

#81

Earlier quoted context omitted.

AI strips a lot of the intention from the work. The artist might want a wooden house with a straw roof and while they design the house they might add framing and other small touches to the house and to the viewer the small decisions could be as important as the choice for a straw roof. When the artist AI generates it they get a house with wooden house with a straw roof and nothing else has any thought or intention. I…

So the artist simply accepts the AI generated wooden house. Might it be possible that he does not? Furthermore, before AI, artists also used assets that came from other creators, accepting them wholesale. No intention, then? The issue I see here is that we're judging the tool, not the artist. It reminds me so much of the digital photography backlash. But here we are.

Digital photography is completely different. The principles do not change whether you’re shooting film or digital. You have to frame the shot. You have to make sure the exposure is what you want. You have to set f-stop. Even if you try to argue that automatic settings on cameras negate all of that, which I don’t agree with but let’s allow it, at the end of the day you have to pick the focal distance and frame, which are the anchor points of the entire photograph. That is not the same as an AI generated facsimile, and whether you’re shooting film or digital does not change that fact.

When use a LLM tool to write, paint, or generate a photograph, you are not actually writing the words, you are not actually physically painting, and you are not framing a photo. It is literally a different process and medium. You can enjoy that, even I have at times, but it’s not the same thing.

Re: WorldClaw Agentic 3D open-world generation at scale

#82

Earlier quoted context omitted.

Even if 99% of generations are trash, why does that matter? Just throw out the failures and keep the good ones, like any writer. Here is the text of the winning story: https://granta.com/the-serpent-in-the-grove/ If you've ever used ChatGPT to generate stories, you will find the style unmistakable. It's cloying to me, but it did manage to fool some stuffy literary judges.

Can you point to some aspects of the style that are typical of ChatGPT? I have never done any story generation and after reading the first couple of paragraphs, I could not tell whether this was AI-generated or not.

AI tends to really like staccato writing, a series of sentences that are too short and simple. It's evident here, although broken up with really long sentences.

Tortured and near-miss metaphors are another thing it likes and it's loaded here. "groaned like a drumskin too tight for the heat" sounds cool but then you think, do drum skins "groan"? And what does heat have to do with it? Or "breath of hills holding their heat like a secret", are they holding the heat or breathing it? At least for me there's the tendency to at first feel like I'm reading something really clever, words that are really tuned and carefully chosen, and then to start feeling like I've been pranked.

Ultimately I don't think there's any way to really be sure though, whether it's AI or just a human but annoying style. This doesn't really read as what GPT would typically drop from a basic writing prompt though, I would guess it's heavily prompted on style or reworked repeatedly.

FWIW though I experiment quite a bit with AI writing for fun, and I think Opus writes pretty nicely in its default voice (prompted mostly for content with little guidance about "how" to write). I'm an engineer so I don't know if real literary connoisseurs would see the slop right through this, but it's coherent, well paced, a little heavy on similes and metaphors but they land unlike the linked passage. Just a sample:

---

Off the port quarter, the darkness lights up orange in a rolling line, gun by gun, and the sound arrives a full second behind the light.

Broadside.

Most of it goes over — a ripping, tearing sound in the air above, like heavy cloth torn by God. Two don’t. One takes the bulwark forward and turns eight feet of oak into a spray of splinters a foot long. One goes into the hull at the waterline abaft the fore chains with a sound you feel through your boot soles, and the sugar barrels in the hold scream as they shift.

A man is down by the foremast, shrieking, and another is not shrieking, and Homer is already running toward both with his hands out.

And Micheal is still aloft.

He’s on the fore topmast crosstrees with the stay gone under him, forty feet up, hanging off a spar that is swinging in a fifteen-degree arc and thumping the mast each time it comes back. You can hear him from the deck. Not screaming. Talking — fast, high, unbroken, reciting the rules to himself. "Touch wood don’t whistle don’t whistle spit to leeward—"

Re: WorldClaw Agentic 3D open-world generation at scale

#83
post #6

In case it's not obvious, this isn't a model, this is python scripts that call out to models (code not available). It's mostly "did you know you can attach an LLM to a PCG system?", but there is one idea here you don't see much of: an image model performs the composition (which image models are really good at), and then you extract the objects into 3d via things like SAM3D before placing them in the world, which is p…

[dead]

It’s close. But “there” is a subjective statement.

Re: WorldClaw Agentic 3D open-world generation at scale

#84
post #69

Earlier quoted context omitted.

LLMs are a PCG system. This is one of my pet peeves, I've been working on a procedural generation constraint system for quite a while (way before LLMs) as a hobby, that came with a domain specific language, where you could specify constraints. Things like the red key must be reachable before the red door, and there should be no path that bypasses the red door, etc. written as a set of constraints. I was working on is…

The trouble is the LLMs non-determinism and the "You're absolutely right, I placed the red key behind the red door rather than in front of it. This was load bearing for the whole dungeon and I blew it" moments. With a hand authored constraint system if the red key is in the wrong place you have a bug in the constraint system that can be deterministically fixed, with a LLM you have to roll the dice on prompting again…

I had a validation script in the background that checked the generated graph - constraint based PCG often fails (it gets into an unsatisfiable state) anyways, and you have to (partially) retry it.

Re: WorldClaw Agentic 3D open-world generation at scale

#86
post #5

Looks impressive, haven't looked at the code yet, I'll at night.

What code, isn’t this just LLM output

Yeah, I saw the example with a link to GitHub and thought there would be a project. Don't like to look at code on my phone though, so I usually wait until I'm on the computer

Re: WorldClaw Agentic 3D open-world generation at scale

#87

It looks impressive. However open worlds are at their best with hand placed details and environmental storytelling. Look at Skyrim/Cyberpunk vs. Starfield where most of the world is proc gen’d. I get that you run this, and then edit it. But the generated villages just aren’t interesting in my opinion. It’s probably great for tencent’s market where you’re mass producing gacha style games, but I don’t think worlds made…

It's at the very least very good starting point, devs should still go in and add mojo to the output

Re: WorldClaw Agentic 3D open-world generation at scale

#88

Earlier quoted context omitted.

Even if 99% of generations are trash, why does that matter? Just throw out the failures and keep the good ones, like any writer. Here is the text of the winning story: https://granta.com/the-serpent-in-the-grove/ If you've ever used ChatGPT to generate stories, you will find the style unmistakable. It's cloying to me, but it did manage to fool some stuffy literary judges.

If 99% generations are trash then it means it sucks at storytelling and is not good at it contary to what was claimed in the comment.

And I think what people don’t appreciate is that when people write/make something, they generally don’t just throw it out there and expect praise. They edit and sculpt it for public release or abandon it. But for some reason when it’s made with an LLM there is this assumption that you need to put it out there because everyone would be so excited to see it. Then when people say it’s not good, everyone starts getting into an argument about AI generated content. I just don’t see the point to this dance anymore. It was interesting for like…the first couple of months maybe.

Alternatively: “Here’s my fragile app I made in a few hours that’s 10x too long and won’t survive iteration but it kind of works. Everyone should thank me.”

I love tinkering, I love messing with things and doing cool projects, but 99% of what I make never goes beyond the room I made it in. Maybe I will show it to a friend when they come over. But ultimately they’re not fit for mass consumption and I don’t expect praise and use.

Re: WorldClaw Agentic 3D open-world generation at scale

#90

Earlier quoted context omitted.

If 99% generations are trash then it means it sucks at storytelling and is not good at it contary to what was claimed in the comment.

And I think what people don’t appreciate is that when people write/make something, they generally don’t just throw it out there and expect praise. They edit and sculpt it for public release or abandon it. But for some reason when it’s made with an LLM there is this assumption that you need to put it out there because everyone would be so excited to see it. Then when people say it’s not good, everyone starts getting i…

Yes, we like tinkering because the process itself is rewarding. The outcome is usually not that important as long as the process is fun. But with AI, the process cannot be fun (because one technically skips it) if one uses it as a way to just spam out content/code. I still say there is a good argument to be made that ai can help in the process of tinkering as well but approach matters a lot and obviously the showcase part of it is not really that important in this context so we most only see short term outcome focused ai projects out in the wild
Post reply on HN