HNHacker News
TopNewBestAskShowJobs

yablak

977 karma · joined January 27, 2015

https://twitter.com/yablak
submissionscomments
yablak··on Don't force your kids to do math
Get a good teacher. They make it fun, or interesting.
yablak··on [dead]
Federal grants are relevant to the interests of many HNers. Why the attack? Chill.
yablak··on After 47 years, OpenVMS gets a package manager (VSP)
It already has a package manager. Emacs.
yablak··on New tools for building agents
Loads fine for me. Maybe because I'm logged in?
yablak··on Subway crime plummets as ridership jumps significantly in congestion pricing era
But...but..but... J hole!
yablak··on Show HN: Pyper – Concurrent Python Made Simple
How does this compare to https://github.com/svenkreiss/pysparkling?
yablak··on Ad Spam in Slack
-- "If you liked reading this post, please follow me on X and check out some of my other posts!"

lol the irony

yablak··on Why does storing 2FA codes in your password manager make sense?
I'm also interested in the answer to this question. Can one separate the cracking of the password and the key?
yablak··on PyTorch Native Architecture Optimization: Torchao
Best way to use tensorflow is by writing models in Jax.
yablak··on Sourcegraph went dark
Sourcegraph search is amazing. I can point to any hash in our repo and search by regex/path regex. Results are instant and in json format. I hacked together a 'cs' script in bash using the sg cli client and some git calls, as I missed Google's cs command since leaving. Works perfectly, faster than ctags/any local indexer.
yablak··on Does Sundar Pichai/Search team know how bad Google search is?
I use Kagi also, but the reason that it's at least as good as Google is that they use the Google search API as a major source of search results.
yablak··on My spiciest take on tech hiring
She
yablak··on Ask HN: Google Search is completely unusable. How do you work around it?
I pay for Kagi.com
yablak··on Modern Git Commands and Features You Should Be Using
Still amazed there's no "git cp". the workarounds involve rebases and aren't practical.
yablak··on Katawa Shoujo: Re-Engineered. Fan-made modernization of the classic visual novel
Katawa Shoujo (translated "Disability Girls") is a bishōjo-style visual novel by Four Leaf Studios that tells the story of a young man and five young women living with varying disabilities. The game uses a traditional text and sprite-based visual novel model with an ADV-style text box running on the Ren'Py visual novel engine. The game is licensed under the Creative Commons CC-BY-NC-ND.

Katawa Shoujo: Re-Engineered offers a battery of accessibility features to accommodate players with disabilities. Using the tools available in modern Ren'Py, we've added quick access to self-voicing, two dyslexic typefaces, a special high contrast text box, and various optimizations to improve functionality with common screenreaders- and that's just the beginning. We're already looking into other ways to make KS:RE even more accessible.

yablak··on Ask HN: Books that gave you different perspective on religion
"The Martyr" by Poul Anderson
yablak··on Ask HN: What underrated open source project deserves more recognition?
https://github.com/NightscoutFoundation/xDrip

xDrip+ is an unofficial and independent Android app which works as data hub and processor between many different devices.

It supports wireless connections to G4, G5, G6, G7, Medtrum A6, Libre via NFC and Bluetooth, 630G, 640G, 670G pumps, CareSens Air and Eversense CGM via companion apps. Bluetooth Glucose Meters such as the Contour Next One, AccuChek Guide, Verio Flex & Diamond Mini as well as devices like the Pendiq 2.0 Insulin Pen.

Heart-rate and step counter data is processed from Android Wear, Garmin, Fitbit and Pebble smart-watches and watch-faces for those that show glucose values and graphs.

On some Android Wear watches, it is possible for the G5 or G6 to talk directly to the watch so it can display values even when out of range of the phone.

The app contains sophisticated charting, customization and data entry features as well as a predictive simulation model.

yablak··on Breakthrough in predicting chaotic outcomes in three body systems
Well, that scifi book aged poorly.
yablak··on Google's winning submission to IWLS2023 Programming Contest
Sadly afaict GitHub offers no viewer for MP4 content.
yablak··on Google's winning submission to IWLS2023 Programming Contest
Google's winning submission to IWLS2023 Programming Contest

The goal of the contest was to synthesize AIGs[1] with the minimal number of two-input and-nodes (AIGs) and two-input xor-and-nodes (XAIGs), for a set of completely-specified multi-output Boolean functions represented using truth tables.

TL;DR: Build the smallest possible circuit that provides a given binary input/output mapping. The solution uses a technique called Circuit Neural Networks and Simulated Annealing.

1. https://en.wikipedia.org/wiki/And-inverter_graph

yablak··on This UN vote shows how completely isolated we are from world opinion
Call them Americans. As in "My fellow Americans.".
yablak··on Translation and accelerated solving of differential equations on GPU platforms
There are two things going on. First, you're right that the vmap should have been done once outside the timing. But equally important, vmap(jit(...)) speed will generally be lower than jit(vmap(...)).

There are many reasons for this. First, the former will loop iterations on the GPU in a serial fashion. Second, the internal jit makes optimization options opaque to Jax. For example, if there's a loop of matmuls inside main, that loop can be converted to a loop of einsums if you vmap first. It can also be fused into sometimes into a bigger operation that doesn't jump control variables back and forth between CPU and GPU between time steps. Between the two you both increase throughput and decrease latency.

I think in Jax, jit(vmap(jit(...))) will also reoptimize the same way as jit(vmap(...)) but I'm not 100% certain.

yablak··on Translation and accelerated solving of differential equations on GPU platforms
Yes, that's right
yablak··on Translation and accelerated solving of differential equations on GPU platforms
Same for pytorch. I don't know enough pyt, but guessing they didn't jit anything.

https://github.com/utkarsh530/GPUODEBenchmarks/blob/ef807198...

yablak··on Translation and accelerated solving of differential equations on GPU platforms
What they should do is build the vmap and jit that, then run timing on calling the resulting function.
yablak··on Translation and accelerated solving of differential equations on GPU platforms
Uhh they time the vmap of the jit on Jax, basically skipping a ton of optimizations,.esp if there is any linear algebra in there. They also include the cost of building the vmap functional. Not a valid comparison.

https://github.com/utkarsh530/GPUODEBenchmarks/blob/ef807198...

yablak··on Ask HN: What sub $200 product improved your 2023
Personal Capital... It's free but belongs on this list: https://www.personalcapital.com/
yablak··on Ask HN: Why do people use password managers?
1password family uses a private key as well as a master password.

Good luck cracking that remotely.

yablak··on Ask HN: What software did you purchase that positively impacted your family life
Also it skips the dumb ads when you search via Kagi.
yablak··on Any other open-source projects like xDrip+?
I recently stumbled upon this OSS project while looking into CGM (Continuous Glucose Monitors). It's extremely feature-full, aggregates data from many types of glucose monitors, and provides open interoperability hooks for other tools. It even uploads the data to Google Health/Fit if you request.

It's a labor of love, created by the Nightscout Foundation - a bunch of folks who wanted to make life easier for family members who have Type 1 diabetes.

It's been developed for years. It's open-source. It's free as in beer. It supports interoperability of data or other additional good-for-the-commons aspect. It has medical or other direct impact on QOL of some group of people.

← PreviousPage 3 of 7Next →