antonmks

GitHub Copilot is now written entirely in Rust, with AI agents doing most of the porting work. The migration cost about $120,000 in AI token usage plus about three weeks of a developer's time. The effort updated the runtime module-by-module until the job was completed, spanning over 135 releases across a 14.5-week time period. 430,000 lines of TypeScript were converted into 800,000 lines of Rust.

There will be a lot of hurt and pain in mathematician's community. It is hard to accept that major discoveries are now just a function of spent token $$.

Absolutely magical drug ! It makes you thinner, younger and now smarter !

So many em dashes...

Definitely birds all the way, with fashionable vertical design.

Spotify and the three main major record labels sue Anna’s Archive for $13trillion for “brazen theft of millions of files containing nearly all of the world’s commercial sound recordings”.

NVIDIA executives allegedly authorized the use of millions of pirated books from Anna's Archive to fuel its AI training. In an expanded class-action lawsuit that cites internal NVIDIA documents, several book authors claim that the trillion-dollar company directly reached out to Anna's Archive, seeking high-speed access to the shadow library data.

Laughing.

A plumber. Very resistant to AI.

It is pretty strange that a country doesn't control what is going in and what is going out. In a small European country I'm most familiar with, everything is checked by customs officers. Dogs, x-rays, customs declarations, import taxes.

I'm experienced in building high-performance data pipelines using Python and JAX, specializing in low-level optimizations like JIT compilation, vectorization, parallelization, and operator fusion. My solutions run seamlessly on CPUs, GPUs, and TPUs, consistently outperforming traditional tools like pandas and polars in both speed and scalability. If you need blazing-fast, scalable data processing leveraging modern hardware and cutting-edge compiler tech, let’s connect!

https://github.com/antonmks https://substack.com/@antonmks

antonmks@gmail.com

Very interesting ! I looked at the repo and it seems that Sirius uses cudf as an engine. So it is not like relational operations were written from scratch. Also, TPCH SF=100 would fit nicely into GPU memory. Would be interesting to see comparisons of something like SF=1000.

How we can use tensor operations—typically used in deep learning—for fast, parallel data processing tasks.

I'm writing a series of tutorials on solving algorithmic leetcode-like problems in a modern, scalable way. I use JAX library for python, so the solutions work on cpus, gpus and tpus. Everything must be vectorized, parallel, just-in-time compiled and differentiable !

https://substack.com/@antonmks

It is not really Postgres, the queries are run on DuckDB. Yeah, DuckDB is really fast for analytical queries.

SEEKING WORK | Worldwide | Remote

Location: Israel

Remote: OK

Willing to relocate: Yes if necessary

Technologies:

- Python/Polars/JAX/CUDA/C++/SQL

- ClickHouse, DuckDB

- Other (AWS, GCP)

- Whisper, STT, TTS

Email: theberlinbird@gmail.com

Database engineer(Python, SQL, High Performance apps) experienced in building voice recognition apps with Whisper on gpu and tpu.

SEEKING WORK

Location: Berlin, Germany

Remote: Yes, any timezone.

Willing to relocate: Depends on location.

Technologies: Python Flutter, Dart, SQL, relational databases, AI, LLM, speech-to-text, text-to_speech

GitHub: https://github.com/antonmks Google Play : https://play.google.com/store/apps/details?id=com.berlin.bir... - app that helps you learn foreign languages.

Email: See my HN profile.

I'm a developer with 20+ years of professional experience. I develop AI, speech-to-text and text to speech solutions for startups and established companies.

Rate : $10/hour

When it comes to writing stories, this model is way behind ChatGpt 3.5

SEEKING WORK

Location: Berlin, Germany

Remote: Yes, any timezone.

Willing to relocate: Depends on location.

Technologies: Flutter, Dart, Python, SQL, AI, LLM, speech-to-text, text-to_speech

GH: https://github.com/antonmks

Email: See my HN profile.

Hi there. I'm a developer with 20+ years of professional experience. I like developing AI, speech-to-text and text to speech solutions for startups and established companies. I can code MVP with advanced AI capabilities for your startup in a relatively short period of time.

It is already too late. The extinction already started. Right now it is insects, frogs and fishes that has got to go. Our time , together with the rest of vertebrates, will come in 60-70 years.

I'm in Belarus now and I can confirm that a developer with fluent English and a few years of experience gets around $400 a month.

"significant reductions in diversity at 415, 322, 300, 145 and 33 Myr ago" - it seems that periods between extinctions events are so large that you can easily fit there hundreds and even thousands of civilizations like ours. So not much to worry about, we will have plenty of opportunities to kill ourselves before a comet does us in.

Ok, here it is : https://gist.github.com/antonmks/17e0c711d41fb07d1b6f3ada3f5...

0.25 cycles/byte using GTX 1080.

SEEKING WORK - Remote only(UTC+3)

I'm a software engineer who specializes in development of applications for gpus. I have experience working with clients in U.S., Europe and Israel.

Technologies: C++, CUDA, NVIDIA libraries (Thrust, ModernGPU), SQL

Github - https://github.com/antonmks/

You can reach me at antonmks@gmail.com

Could be a good idea for Windows 10 - blocking all ads by default, including those delivered through browsers. It could seriously affect and even kill many current and potential Microsoft competitors like Google and Facebook.

Sorry if you are misled :-) The program does parses selected fields, there are strings-to-binaries procedures.

I tested this approach on multiterabyte files, take a look at my alenka project, it uses the same method to load large CSV files into databases. It just have to be done in chunks.

The program compiles fine, that pull request was referring to incorrect earlier version of test.cu file.

Test it for yourself, see if you get similar results.

Thrust CUDA calls are synchronous to each other. You can add an explicit synchronization call cudaDeviceSynchronize() and there won't be a difference in results.

I used average time for multiple runs, both GPU and CPU benchmarks ran from memory.

It should be 750MB, not 750GB. There is no way a single hard drive can read 750GB in 14 seconds.

Yes, CPU version can be accelerated as well, depending on availability of SSE extensions.