From Julia to Rust: a differentiable tensor stack for scientific computing(tensor4all.org)

47 pointsby postflopclarity2 days ago4 comments

jarbusa day ago
really 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.
- runarberga day ago
  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).
  - hogehoge51a day ago
    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?
    runarberg15 hours ago
    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.
    hogehoge5114 hours ago
    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.
nestorDa day ago
I 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...
Archit3cha day ago
Metal 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. :(
QuadmasterXLIIa day ago
Disconcerting 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?
- hogehoge51a day ago
  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.
  - QuadmasterXLII13 hours ago
    There's no larger philosophical point here, just an empirical observation that when I trust an AI generated readme I reliably get burned, in a way that I typically don't get burned by e.g. an AI generated svg viewer or batch download script.
    hogehoge5113 hours ago
    Makes sense, that's a fair heuristic.
    TBH - I weigh it higher as it's from a University in Saitama, a place I have a strong connection to and don't see here often. An irrational reason to ignore the AI smell, but why not.