HNHacker News
TopNewBestAskShowJobs

m3at

1,949 karma · joined April 26, 2018

ML specialist based in Tokyo. If you like to talk about AI or coding, I'll buy you a coffee!

contact at paulw dot tokyo

blog: https://paulw.tokyo

twitter: @PaulWTokyo

submissionscomments
m3at··on Ask HN: Who is hiring? (July 2023)
Good pitch and I appreciate that your landing page itself is in the spirit of what you do!

> heavy R&D in AI / LLM / GPT

Can you expand on your plans in the AI space? I'm full stack for everything ML related with 7 years in the field, but quite limited frontend wise! If that's something that can be valuable to your team I can get in touch

m3at··on Ask HN: Could you share your personal blog here?
https://paulw.tokyo/

Few posts a year, mainly about machine learning

m3at··on Ask HN: Who is hiring? (June 2023)
I think it's hard to come up with a regex that'll cover everything, but that's the kind of things GPT is good for. I made put some scripts here, works well enough: https://github.com/m3at/hn_jobs_gpt_etl
m3at··on Ask HN: Who is hiring? (June 2023)
Very interested, will send you an email shortly
m3at··on Let us serve you, but don't bring us down
You can find some demos linked on github [1], the simplest being: https://demo.mcaptcha.org/widget/?sitekey=pHy0AktWyOKuxZDzFf...

[1] https://github.com/mCaptcha/mCaptcha#client-side-widget

m3at··on Ask HN: Side project of more than $2k monthly revenue? what's your project?
I just tried well word, unlike a sibling comment I think it's fun :)

Though I have to agree that it's a bit unclear how to proceed at first. Maybe make a link to the tutorial more easily accessible, or even auto-suggested on first visit?

m3at··on Run CLIP on iPhone to search photos
Nice work! I've made a similar demo as a web app [1] and obtained a few orders of magnitude faster performance, in the 50ms for 1.6M images on single laptop core. I've limited experience with coreml but there's a few tricks that could probably help you too? If you feel like hacking on it together shoot me an email :) (in profile)

[1] https://paulw.tokyo/post/real-time-semantic-search-demo/

m3at··on Orion Browser
> By default Brave will not block first Party ads which are actually part of the sites you visit

https://support.brave.com/hc/en-us/articles/360022806212-How...

m3at··on How do I draw a pair of buttocks? (2014)
The other way around
m3at··on Do simpler machine learning models exist and how can we find them?
Will do if I have feedback :)
m3at··on Do simpler machine learning models exist and how can we find them?
Now I realise that I _somehow_ totally missed your link to Page's work right in the README…

Thanks for the detailed comment, I definitely will dive into your script soon! One early suggestion, you might want to try torch dynamo [1], anecdotally I had good speedups (~20%) on some image models; though not sure how significant the impact might be at this (relatively) small scale.

[1] https://pytorch.org/docs/master/dynamo/

m3at··on Do simpler machine learning models exist and how can we find them?
Nice experiment! Not to minimize your work, in fact some ideas might be complementary to yours, a couple years ago there were already some approaches reaching 30s on a single V100:

https://myrtle.ai/learn/how-to-train-your-resnet-8-bag-of-tr...

m3at··on OpenAI quietly launched Whisper V2 in a GitHub commit
Agreed. It's maybe like Linux versioning scheme, where 6.0 is just the one that comes after 5.19:

> The major version number is incremented when the number after the dot starts looking "too big." There is literally no other reason. [1]

[1] https://www.kernel.org/category/releases.html

m3at··on Stable Diffusion with Core ML on Apple Silicon
That's done in professional contexts, when you only care about inference onnxruntime does the job well (including for coreml [1]).

I imagine that here apple wants to highlight a more research/interactive use, for example to allow fine tuning SD on a few samples from a particular domain (a popular customization).

[1] https://onnxruntime.ai/docs/execution-providers/CoreML-Execu...

m3at··on Housing in Tokyo is cheaper and more spacious than people think
A peculiarity of renting in Japan (not mentioned in the article): you often need to advance a substantial amount of money upon starting a lease. It varies but a rule of thumb is about 5 months of rent. Part of it is advanced rent, but some is non-recoverable (key money, brokerage fee, insurance, cleaning fee...). Contracts are often two years, and at the end of it you typically have a renewal fee as well, at around a month of rent.
m3at··on Falcon Heavy Launch [video]
This is a launch of falcon heavy, there's only been a handful (unlike single falcon 9, which are more common and might be the "random SpaceX launch" you had in mind). It's also a live stream instead of a news article. But other space news events are interesting too of course
m3at··on Adobe to acquire Figma for $20B
I think you misread, the parent agree with you that Figma created value, the issue is the acquisition by Adobe that might stop or slow this.
m3at··on Apple surpassed $3.5B in annual revenue from its ad network
Link to the full article: https://www.youngmoney.co/p/apples-diabolical-plan-control-a...
m3at··on Ask HN: In 2022, what is the proper way to get into machine/deep learning?
There's so much resources it's hard to pick… If you're already studying CS and like to learn by building, I'd recommend to do that for ML too!

Pick a real problem, try to build a ML solution for it and while doing so keep a list of things you'd like to dig deeper into. Then go back to that list and pick one item to study, and iterate.

Happy to have a chat and give you specific pointers if you'd like (email in profile), I got my master in ML in 2016 and applied it in the industry since.

m3at··on Kagi search and Orion browser enter public beta
You can with the !bang syntax like in ddg, example !g for google (then keep the rest of your query)
m3at··on Kagi search and Orion browser enter public beta
I'm in the same boat and will pay for Kagi, though I have a different interaction with a search engine and thought you might like the counterpoint. Notably, I don't use Google for most of what you described:

Translation: I use DeepL instead (google is doing poorly on Asian languages)

Conversion: not having much need for imperial but I use this for all unit conversion (though usually directly in my terminal): https://insect.sh/

Wikipedia: if I type Wikipedia+something my browser automatically use the wiki engine (chrome can do that too)

Word definition: on OSX I long press on the trackpad and the definition pops up (across languages, translation too for single words)

Lyrics: Spotify show them now! (admittedly that's a recent feature)

m3at··on Using a "proper" camera as a webcam
Hmm maybe vidgear can do it? I'm unsure about the latency you'll get out of it though:

https://abhitronix.github.io/vidgear/v0.2.5-stable/bonus/ref...

m3at··on Using a "proper" camera as a webcam
You can directly with gstreamer, but it's a bit of a hassle to set up. You could try this python convenience library: https://git.aweirdimagination.net/perelman/minimal-webrtc-gs...
m3at··on Nvidia Research Turns 2D Photos into 3D Scenes
This is great, and the paper+codebase they're referring to (but not linking, here [1]) is neat too.

The research is moving fast though, so if you want something almost as fast without specialized CUDA kernels (just plain pytorch) you're in luck: https://github.com/apchenstu/TensoRF

As a bonus you also get a more compact representation of the scene.

[1] https://github.com/NVlabs/instant-ngp

m3at··on Ask HN: What ML platform are you using?
To support your point with data, here is a graph of usage of TF vs PyTorch in papers over time: https://horace.io/pytorch-vs-tensorflow/
m3at··on The Nimatron: first 'video game' debuted in 1939
Others are pointing out that it's hard to call this a video game (agreed), but I never heard of the nimatron and it was interesting. So thanks for sharing anyway!
m3at··on WebVM: Server-less x86 virtual machines in the browser
Sibling comments have mentioned web torrents, you could also look at IPFS for something related to what you're describing (without the abuse) https://ipfs.io/
m3at··on U.S. surgeons transplant pig heart into human patient
A documentary called South Park showed just that in this episode: https://en.m.wikipedia.org/wiki/Raising_the_Bar_(South_Park)

/s

m3at··on U.S. surgeons transplant pig heart into human patient
For even more of a tangent, there is a South Park episode (S16E9) with a heart transplant from pig to human

https://en.m.wikipedia.org/wiki/Raising_the_Bar_(South_Park)

m3at··on New York City bans natural gas in new buildings
Yes, I remember dang making a similar remark (can't find the link atm though)
← PreviousPage 2 of 9Next →