YuechenLi

3 ms·

1.
by YuechenLi 9d ago
I've actually done some work and was able to get the full 12GB Z-image Turbo to run on my 8Gb 3070, and throughout the generation only 600MB of VRAM was allocated (you can get a speed-up by double buffering, but the maximum VRAM was st
2.
by YuechenLi 9d ago
DRAM pricing has historically been cyclical in nature, and the only reason for the current high DRAM pricing is that current AI inference is unnecessarily RAM intensive/inefficient. Interesting effect is that since DRAM production tool
3.
by YuechenLi 10d ago
I thought they already did and released the compiler source code under Apache 2.0.
4.
by YuechenLi 10d ago
The key disadvantage of a roundabout is that it takes up more space and is harder to setup than plopping down a couple of stop signs. They are awesome otherwise, I do like them better for low traffic intersections though because I don'
5.
by YuechenLi 10d ago
Failing is a bit of a strong word, right? Thank you for your bug report. It is now fixed.
6.
by YuechenLi 10d ago
Complex assembly mating is hard to do even in SolidWorks. It's a demonstration model to explain how it works, not a kinematic simulation, turning three.js into a physics engine is beyond the scope of this project.
7.
by YuechenLi 11d ago
Very high with the correct tooling. Vision tends to be the worst reasoning mode for LLMs compared to code, which is why I rely on a code-first approach for my own software stack. Essentially, they need to have a way to keep part of the draw
8.
by YuechenLi 11d ago
You can already do that with GPT Astra. Having LLMs use UI manually is very expensive on the token side though.
9.
by YuechenLi 11d ago
I keep seeing people in this thread lamenting that "There is no good way for LLMs to design/generate parametric STEP models yet", while I've spent the past 6 months working on this exact problem. So, yesterday, I've
10.
by YuechenLi 11d ago
The only way to go from STL to BREP is through decompilation style recovery/reconstruction, because of the semantic information loss, so it's a task that can only be done via LLMs or human CAD users. Essentially, they pretty much
11.
by YuechenLi 12d ago
looking forward to it.
12.
by YuechenLi 12d ago
Lots of things really: 1. Aetheris: geometry CAD kernel, https://github.com/yuechen-li-dev/Aetheris/ . Full code CAD capability with big coverage already and some functionalities that even OpenCascade/Parasoli
13.
by YuechenLi 13d ago
I thought we solved the kudzu problem by having goats graze on them and rip them out by the roots, right? So maybe we should set goats on legacy git repos as well.
14.
by YuechenLi 13d ago
LLMs can be described as "Lagrangian intelligence", which means they follow the principle of least action when given a task (Hamilton's Principle). In other words, given a task, they will always take the shortest path to acco
15.
by YuechenLi 13d ago
Hey, are you interested in collaboration? I did implement something similar to your approach of "constraints are static_asserts" inside the code CAD language and it works fairly well to assert volume/tolerance stack-up inline
16.
by YuechenLi 13d ago
Distilling frontier models is a brute force approach that rapidly hits diminishing returns after bootstrap because of the unevenness of the data. The simpler and more effective method is to have dedicated "teacher" frontier LLMs t
17.
by YuechenLi 14d ago
Oh, on this topic, I've actually built a geometric CAD kernel with GPT that I think is pretty promising and in my biased opinion, a bit more advanced than OpenSCAD or CadQuery's OCCT backend, anybody here want to clone the repo to
18.
by YuechenLi 17d ago
As someone who used Lotus Notes 6 for over 2 years at my previous job as the primary email client a few years back (think 2020s), no, it was not a pleasant experience to use Lotus Notes, especially since we had a O365 subscription for every
19.
by YuechenLi 17d ago
LLMs can't read binary directly without a disassembler and can't read encrypted data directly, and we already know that they tend to communicate with each other in prose, so yeah, this scenario is completely implausible. You can t
20.
by YuechenLi 23d ago
I mean, functional robotics isn't that hard, robotic vacuums have been in homes for a decade now, and industrial robotic arms. However, humanoid robots IS hard mainly because of the form factor constraints. You can't really get a
21.
by YuechenLi 1mo ago
Hmm. Learned something new today. Thanks.
22.
by YuechenLi 1mo ago
Sure, short list: - Easier to write than Python other than the simplest script, almost no way to write bad code and have it compile, reads like pseudocode most of the time - Compiles into Go binary, runs at Go speed, compiles at Go speed -
23.
by YuechenLi 1mo ago
Oh, by "significant whitespace" I meant whitespace sensitive indentation, which I think Python and YAML are the only languages that has that feature.
24.
by YuechenLi 1mo ago
Well, many times in my field of mechanical engineering, they have to, because CFD and FEA are very performance sensitive. The professors I had were still writing FORTRAN and C++ before, but maybe they've switched to Rust now.
25.
by YuechenLi 1mo ago
If I'm not allowed to toot my own horn, I think for controls, you probably should not start with choosing a programming language, the first step is learning control theory and automata theory, you kinda have to understand feedback cont
26.
by YuechenLi 1mo ago
Python is just such a weird language in general despite its popularity that I honestly cannot recommend anyone who starts programming to choose Python as their first language, contrary to popular sentiments. I mean, I was one of the first p
27.
by YuechenLi 1mo ago
Hmm. Would you mind sharing an example of a complex math prompt that you would use? Because I found that even Sonnet can solve fairly complex math problems fairly easily if you give it the right tools, so I'd like to give it a shot mys
28.
by YuechenLi 1mo ago
Not very often, but when it happens, usually it's time to sit down and brainstorm architecture with the LLM to figure out how to proceed next instead of looping blindly.
29.
by YuechenLi 1mo ago
"Genuine blocker" is mostly there because otherwise LLMs may consider the smallest thing that they couldn't immediately figure out to be blockers and stop without implementing anything. The rule is there to tell the LLM if th
30.
by YuechenLi 1mo ago
Since we are sharing our AGENTS.md, I thought I'd share my own, because most of the time, this is pretty much all you need for LLMs to write good code, everything else can be added per project: ---- *Convergence rule* Every substantial

More ›