I just posted this somewhere else -- but overall big fan of these text to cad rigs as projects.
Obligatory mention of https://zoo.dev/ who went to extreme lengths on this.
I will say I explored this reasonably deeply and came away with the conclusion that even though we have OpenSCAD and all these examples, LLMs are still very weak at spatial reasoning compared to diffusion models.
You can do all sorts of tricks like have a parts library to get around this and do physics checks but another inconvenient truth is whenever you design a complex assembly, every change to that part needs to be aware of the other parts in the design -- thus you need a global part-aware editing capability from diffusion.
That's getting solved already in china leading labs, and bottlenecked by the lack of good training data, which china is solving with mass labor.
This will be solved overseas first before we will in the US.
p.s. I am not affiliated with zoo or any of these other things FYI was just very curious about this whole area
> LLMs are still very weak at spatial reasoning compared to diffusion models
Don't know what diffusion model can do, but 100% agree with the "LLMS are very weak at spatial reasoning" comment.
I build a rather complex blueprint-image-to-3D-brep-model a couple of months back using codex ... ugh the damn thing has really no idea where things are in space, something a 3 year old figures out instinctively.
It did end up saving some time as compared to modeling the object myself in a CAD package, but there was so many completely obvious thing I had to explain ... very hard to believe when compared to what codex can pull of with code.
I've been watching the space as well, waiting for the day I can stop fiddling with widgets and just tell the damn thing about the shapes I want and the ways in which they will move. Alas, we're far from that yet.
> That's getting solved already in china leading labs
Care to drop a bit of info as a follow up to this claim? Curious!
Obligatory mention of https://zoo.dev/ the leader in this space.
I will say I explored this reasonably deeply and came away with the conclusion that even though we have OpenSCAD and all these examples, LLMs are still very weak at spatial reasoning compared to diffusion models.
You can do all sorts of tricks like have a parts library to get around this and do physics checks but another inconvenient truth is whenever you design a complex assembly, every change to that part needs to be aware of the other parts in the design -- thus you need a global part-aware editing capability from diffusion.
That's getting solved already in china leading labs, and bottlenecked by the lack of good training data, which china is solving with mass labor.
This will be solved overseas first before we will in the US.
Ehh, I think GP was saying it doesn’t fit with the tone of a message board reply, so it seems cut and paste from some outside place. Either a press release, a review snippet, or AI.
One would imagine they are broadly similar; but that's off the assumption that codebases are similar as well.
Migrations between versions can have big variance largely as a function of the parent codebase and not the dependency change. A simple example of this would be a supported node version bump. It's common to lose support for older node runtimes with new dependency versions, but migrating the parent codebase may require large custom efforts like changing module systems.
This was created by Zach Latta, who runs an awesome non profit called Hack Club that produces some of the top high school technical talent through community coding clubs. Highly encourage you donate to Hack Club!
Hack Club is amazing! I've gotten so many amazing opportunities to build stuff and help out with events. I got to build and run waka.hackclub.com (selfhosted wakatime backend) and at its peak we got over 21 thousand kids tracking over a quarter million hours in the software i built. Crazy experience. Getting to talk and interact with other techy teens is the best part about it though; we have a massive slack which generates the best techy conversations I typically have on a day to day basis.
I run a sw supply chain company (fossa.com) -- agree that there's a lot of low hanging gains like inventory still around. There is a shocking amount of very basic but invisible surface area that leads to downstream attack vectors.
From a company's PoV -- I think you'd have to just assume all 3rd party code is popped and install some kind of control step given that assumption. I like the idea of reviewing all 3rd party code as if its your own which is now possible with some scalable code review tools.
Those projects seem to devolve into a boil the ocean style projects and tend to be viewed as intractable and thus ignorable.
In the days everything was http I use to set a proxy variable and have the proxy save all downloaded assets to compair later, today I would probably blacklist the public CAs and do an intercept, just for the data of what is grabbing what.
Fedramp was defunded and is moving forward with a GOA style agile model. If you have the resources I would highly encourage you to participate in conversations.
The timelines are tight and they are trying to move fast, so look into their GitHub discussions and see if you can move it forward.
There is a chance to make real changes but they need feedback now.
+1, I think you have to assume owned as well and start defending from there. Companies like edera are betting on that, but sandbox isn't panacea, you really need some way to know expected behavior.
Very interesting research -- possibilities are endless for personalization, rigging, character animation. Especially interesting how few frames it takes to accomplish this.
Obligatory mention of https://zoo.dev/ who went to extreme lengths on this.
I will say I explored this reasonably deeply and came away with the conclusion that even though we have OpenSCAD and all these examples, LLMs are still very weak at spatial reasoning compared to diffusion models.
You can do all sorts of tricks like have a parts library to get around this and do physics checks but another inconvenient truth is whenever you design a complex assembly, every change to that part needs to be aware of the other parts in the design -- thus you need a global part-aware editing capability from diffusion.
That's getting solved already in china leading labs, and bottlenecked by the lack of good training data, which china is solving with mass labor.
This will be solved overseas first before we will in the US.
p.s. I am not affiliated with zoo or any of these other things FYI was just very curious about this whole area
reply