From Julia to Rust: a differentiable tensor stack for scientific computing
1–10 of 13 posts
Re: From Julia to Rust: a differentiable tensor stack for scientific computing
#2really interesting perspective, as an ex-julia user, can't really argue with the main points. I will say that julia is delightful to use and code in, whereas the article's main point is that rust becomes bearable once you don't have to code yourself, haha.
Re: From Julia to Rust: a differentiable tensor stack for scientific computing
#3Disconcerting to see at the end that the blog post is generated. The genre is decidedly “use-my-thing Readme.md” and all current gen LLMs by default jump to shameless lies when they detect that they are writing such a Readme, although they are perfectly capable of working out the truth in a genre like “essay question on which you will be graded by academic standards.”
Are the human “coauthors” lying if I hypothetically go to the crate looking for the promised xla backend and find //TODO implement this?
Re: From Julia to Rust: a differentiable tensor stack for scientific computing
#4really interesting perspective, as an ex-julia user, can't really argue with the main points. I will say that julia is delightful to use and code in, whereas the article's main point is that rust becomes bearable once you don't have to code yourself, haha.
I‘m not reading TFA because it is AI generated, but I recently returned to Julia to prototype an Artificial Neural Network, and oh boy did I miss this language. If I ever decide to continue with the project I am eventually going to have to rewrite in Rust though (to use the WebAssembly target).
Re: From Julia to Rust: a differentiable tensor stack for scientific computing
#5Disconcerting to see at the end that the blog post is generated. The genre is decidedly “use-my-thing Readme.md” and all current gen LLMs by default jump to shameless lies when they detect that they are writing such a Readme, although they are perfectly capable of working out the truth in a genre like “essay question on which you will be graded by academic standards.” Are the human “coauthors” lying if I hypothetical…
Why?
Firstly, it clearly states it was written collaboratively. The usual division of labour splits the task into ideas and words, human the former, agent the latter. Is this a problem? I was interested in understanding the author’s ideas; consuming the words is an unfortunate chore.
Secondly, this was published in three languages. Perhaps you missed the 日本語 and 中文, or dismissed them as agent noise? To reinforce the human/agent demarcation: without agentic AI, it would probably be quite the chore to post all three languages. The English is not written the way a non-native speaker would write it. I assume the author conversed with the agent in their native language to describe the content and conveniently got three sets of generated words expressing their ideas.
I really struggle to understand the conflation of communication-channel coding (words) and authorship.
Re: From Julia to Rust: a differentiable tensor stack for scientific computing
#6really interesting perspective, as an ex-julia user, can't really argue with the main points. I will say that julia is delightful to use and code in, whereas the article's main point is that rust becomes bearable once you don't have to code yourself, haha.
I‘m not reading TFA because it is AI generated, but I recently returned to Julia to prototype an Artificial Neural Network, and oh boy did I miss this language. If I ever decide to continue with the project I am eventually going to have to rewrite in Rust though (to use the WebAssembly target).
Which version of TFA would you read if it was not generated? would you be comfortable reading the 日本語版 if it was human authored?
Re: From Julia to Rust: a differentiable tensor stack for scientific computing
#7I have wanted a JAX-like (XLA backed) Rust tensor library for ever!
Plus it looks like they have support for dynamic shapes: https://tensor4all.org/tenferro-rs/design/dynamic-symbolic-s...
Re: From Julia to Rust: a differentiable tensor stack for scientific computing
#8Metal only supported via wgpu due to pulling in CubeCL as a dependency is... not ideal.
If you need to extend it, you'll have to write custom Autodiff rules with the ergonomics of a Rust front-end and a JAX-like backend. :(
Re: From Julia to Rust: a differentiable tensor stack for scientific computing
#9Earlier quoted context omitted.
I‘m not reading TFA because it is AI generated, but I recently returned to Julia to prototype an Artificial Neural Network, and oh boy did I miss this language. If I ever decide to continue with the project I am eventually going to have to rewrite in Rust though (to use the WebAssembly target).
Which version of TFA would you read if it was not generated? would you be comfortable reading the 日本語版 if it was human authored?
As a matter of fact yes. I much prefer to machine translate TFA my self if it sounds interesting enough. But in this case I wouldn’t even do that because the 日本語版 also has this disclaimer:
> 開示: この記事は AI コーディングエージェントと著者が共同で書いた。仕上げにあたっては、本文で説明したのと同じ、人間が検証する工程を通している。
My Japanese is not good enough that I would be able to detect the チャッピー構文 my self, so I thank the authors for putting it there, and saving me the embarrassment of reading something that the authors didn’t even bother to write.
Re: From Julia to Rust: a differentiable tensor stack for scientific computing
#10Earlier quoted context omitted.
Which version of TFA would you read if it was not generated? would you be comfortable reading the 日本語版 if it was human authored?
As a matter of fact yes. I much prefer to machine translate TFA my self if it sounds interesting enough. But in this case I wouldn’t even do that because the 日本語版 also has this disclaimer: > 開示: この記事は AI コーディングエージェントと著者が共同で書いた。仕上げにあたっては、本文で説明したのと同じ、人間が検証する工程を通している。 My Japanese is not good enough that I would be able to detect the チャッピー構文 my self, so I thank the authors for putting it there, and saving me the embarras…
I would challenge the asserion that there was no effort in writing it. Basic information theory sugests the bits in content needs to come from somewhere. The more the bits diverge from what is in the training corpus, the more effort the human author needs to apply. Unless people also think there is a special LLM bit that differs from human bits of information. It also suggests a lack of appreciation for the effort in delegating any creative work, to humans or agents, and still getting what you asked for.
Regardless of agentic provenance, i liked the chance to read something from Saitama university here. Saitama's a nice place.