tzhenghao

15 ms·

31.
Dan Abramov The wet codebase (2019) [video](youtube.com)
2 points by tzhenghao 3y ago|0 comments
32.
by tzhenghao 3y ago
Yeah, I was just chatting with another principal eng friend recently where maybe a whole new generation of programmers may be several degrees away from topics like operating systems, compilers, computer security and networking. Once in a wh
33.
The disappointingly ongoing success of WvDial (2009)(apenwarr.ca)
65 points by tzhenghao 3y ago|37 comments
34.
John Carmack on Idea Generation (2016)(amasad.me)
3 points by tzhenghao 3y ago|0 comments
35.
by tzhenghao 3y ago
"Show me the incentive, I'll show you the outcome." - Charlie Munger
36.
by tzhenghao 3y ago
Aside from the many software application examples on this thread, have you also considered how they also pushed innovation in the semiconductor industry that in turn contributed to non-obvious advancements in non-Big Tech sectors, from part
37.
Do LLMs eliminate the need for programming languages?(modular.com)
1 points by tzhenghao 3y ago|0 comments
38.
Show HN: Visualize and optimize your PyTorch models on Cellulose(cellulose.ai)
1 points by tzhenghao 3y ago|0 comments
39.
A Review of Sparse Expert Models in Deep Learning (2022)(arxiv.org)
2 points by tzhenghao 3y ago|0 comments
40.
by tzhenghao 3y ago
Good point. "AI accelerators" is a thing now that competes directly with "GPUs used for AI as its sole purpose".
41.
by tzhenghao 3y ago
Eh, time will tell. Nvidia is killing it because they have "semi decent" software toolchains like CUDA when just about hardware player botched software. That said, a lot of interesting development on the XLA and PyTorch 2.0 sides
42.
by tzhenghao 3y ago
I remember working with several ML teams on specifically inference performance (latency, memory usage etc.), and it's not a surprise to see some object detection performance variance depending on a scene. Sometimes, even capping the mo
43.
by tzhenghao 3y ago
Amazing speech. I love that he touched on "the CUDA bet". How their early adopters in the sciences (particle physics, fluid dynamics etc.) eventually made it to the larger AI boom. Many of my former teammates came from Lawrence Li
44.
by tzhenghao 3y ago
Interesting! Thanks for sharing, and definitely quite surprised how big that bandwidth premium is [1] [1] - https://blog.cloudflare.com/bandwidth-costs-around-the-world
45.
Ask HN: Why is there a cloud pricing premium in APAC vs. US or EU?
2 points by tzhenghao 3y ago|2 comments
46.
by tzhenghao 3y ago
This. Modular and OctoML are building on top of MLIR and TVM respectively. > It's pretty hard to really beat NVIDIA for developer support though, they've invested a lot of work into the CUDA ecosystem over the years and it show
47.
by tzhenghao 3y ago
Not surprising lol. This was also the experience I had while experimenting with MLIR approximately 3 years ago. You'd need to git checkout a very specific commit and then even change some flags in code to have a successful build. I&#x2
48.
by tzhenghao 3y ago
I think AMD is going down a different path, ie. ROCm then partnering with ML frameworks further up the stack for first class support. https://pytorch.org/blog/pytorch-for-amd-rocm-platform-now-a...
49.
by tzhenghao 3y ago
> I think the only way to start an AI chip company is to start with the software. The computing in ML is not general purpose computing. 95% of models in use today (including LLMs and image generation) have all their compute and memory ac
50.
Show HN: Cellulose – a tool to improve inference performance of ML models(cellulose.ai)
11 points by tzhenghao 3y ago|0 comments
51.
by tzhenghao 3y ago
I've used tldr ( https://github.com/tldr-pages/tldr ) if I'm a beginner to a specific tool, then read up the manpages and/or SO if I need more flags.
52.
Which GPU(s) to Get for Deep Learning(timdettmers.com)
2 points by tzhenghao 3y ago|0 comments
53.
The Economics of Software Maintenance (2020)(outspeaking.com)
2 points by tzhenghao 3y ago|0 comments
54.
by tzhenghao 3y ago
This. And password managers can do more by constantly checking latest account breaches like on https://haveibeenpwned.com then flagging it to their customers to rotate their credentials.
55.
by tzhenghao 3y ago
I have over 300 on 1Password, some with more than 5 different accounts from the same site like Google and Slack. I also lean heavily on the 1Password extensions to help on the retrieval part. It's incredibly insane but I manage it by a
56.
by tzhenghao 3y ago
Tragedy of the commons - only need a few bad actors to ruin it all for us. Almost all distributors face this problem, from Docker Hub to PyPI. This also reminded me of official Postgres Docker image running a cryptominer in the background [
57.
by tzhenghao 3y ago
Just a little push back here, I think you strike on the right theme where a programming language could fill this gap. However, I wonder if new domain specific languages will eventually be the more elegant solution. Think Modular's Mojo
58.
by tzhenghao 3y ago
This. Just to riff off an example, a lot of the APIs in common DL frameworks like PyTorch revolve around numpy or pickle formats. These are Python first semantics.
59.
Study finds higher cancer rates among U.S. military airmen and ground crews(pbs.org)
44 points by tzhenghao 4y ago|15 comments
60.
by tzhenghao 4y ago
Good points, but I'd push back just a little. > Sure, the actual implementations can be tricky, but a single engineer (trained in it) can get that done for a specific hardware in a couple months. I want to agree with you on this, bu

More ›