HNHacker News
TopNewBestAskShowJobs

artemisart

328 karma · joined March 6, 2016

submissionscomments
artemisart··on Qwen3: Think deeper, act faster
ChatGPT free gets it right without reasoning mode (still explained some steps) https://chatgpt.com/share/6810bc66-5e78-8001-b984-e4f71ee423...
artemisart··on Lossless LLM compression for efficient GPU inference via dynamic-length float
The first sentence of the introduction ends with "we introduce Dynamic-Length Float (DFloat11), a lossless compression framework that reduces LLM size by 30% while preserving outputs that are bit-for-bit identical to the original model" so yes it's lossless.
artemisart··on OpenAI releases image generation in the API
That was the joke.
artemisart··on Civilization VII only needs 20 GB of free disk space
There is absolutely some bloat in 100gb+ games but also come on, graphics are much better now compared to 15 years old games. Textures + video cinematic + models weight a lot.
artemisart··on Sora is here
Yes, the idea works and was explored with dreambooth/textual inversion for image diffusion models.

https://dreambooth.github.io/ https://textual-inversion.github.io/

artemisart··on State of Python 3.13 performance: Free-threading
This depends a lot on your domain, e.g. pypy is not compatible with pytorch or tensorflow so DL is out of the picture.
artemisart··on Computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku
Thanks! I was waiting for your benchmarks. Do you plan to test haiku 3.5 too? It would be nice to show API prices needed to run the whole benchmark too to have a better idea of how many internal tokens o1 models consume.
artemisart··on AI PCs Aren't Good at AI: The CPU Beats the NPU
Important precision: the async part is absolutely not python specific, but comes from CUDA, indeed for performance, and you will have to use cuda events too in C++ to properly time it.

For ONNX the runtimes I know of are synchronous as we don't do each operation individually but whole models at once, there is no need for async, the timings should be correct.

artemisart··on AI chipmaker Cerebras files for IPO
Correction: it's 8x the TFLOPS of a DGX (8 H100), not 1 H100. But it's true that if it stays at $3M it's probably too much and I don't think the memory bottleneck on gpus is large enough to justify this price/performance.
artemisart··on Our container platform is in production. It has GPUs. Here's an early look
I can't find any GPU capabilities on their website and it seems to be 100% focused on game backend hosting, not general workers. Do you have more information?
artemisart··on GIL Become Optional in Python 3.13
I don't really understand such critics which I often see for the python 2 to 3 transition, the OS didn't want to change `python` to be v3 by default which is understandable for backward compatibility, and the fix is simple and immediate with e.g. the python-is-python3 apt package, or an alias python=python3 in the shell, or a symlink in PATH, etc.
artemisart··on Amazon's exabyte-scale migration from Apache Spark to Ray on EC2
That's also $41k / 26k / 13k per day or $1.2M / 767k / 375k per month!
artemisart··on Call of Duty: Warzone Caldera Data Set for Academic Use
Is it really a common hack? I would have guessed teleportation is the easiest to detect server-side, or impossible from the start as the server is authoritative (clients sends inputs, the server computes the positions and any important change, sends them back to clients, clients cannot hack their movement).
artemisart··on "GitHub" Is Starting to Feel Like Legacy Software
Is this (Google Code Search, https://source.chromium.org/) open-source?
artemisart··on Ask HN: Which LLMs can run locally on most consumer computers
No, the Nvidia 4070 Ti has much higher performance, TOPS is for integer operations, the 4070 Ti has ~40 float32 TFLOPS and 641 TOPS https://www.nvidia.com/fr-fr/geforce/graphics-cards/40-serie... (which I would say would be peak TOPS for int4 operations, comparing it to the 4080 datasheet, and a bit more than half that for int8 operations) https://images.nvidia.com/aem-dam/Solutions/geforce/ada/nvid... page 34. I did not find the datasheet for 4070 Ti.
artemisart··on JPEG XL and the Pareto Front
About 3/4 of the article? There is a lot of analysis on lossy compression too after the lossless part.
artemisart··on Details on Xiaomi EV
The quote above made it a bit unclear, they measured it at 1MB vs 52kB per device:

> When idle, Android sends roughly 1MB of data to Google every 12 hours, compared with iOS sending Apple about 52KB over the same period. In the US alone, Android collectively gathers about 1.3TB of data every 12 hours. During the same period, iOS collects about 5.8GB.

artemisart··on GitHub Actions could be so much better
I agree, but of course all CI vendors build all their documentation and tutorials and 'best practices' 100% on the first option for lock-in and to get you to use more of their ecosystem, like expensive caching and parallel runners. Many github actions and circleci orbs could be replaced by few lines of shell script.

Independent tutorials unfortunately fall in the same bucket as they first look at official documentation to try to follow so-called best practices or just try to get their things working, and I would say also because shell scripts will seem more hacky for many people -unfairly-.

artemisart··on Fish – A friendly interactive shell
Just rewrite them once with fish_add_path? https://fishshell.com/docs/current/tutorial.html#path Or maybe use https://github.com/edc/bass
artemisart··on Blender 3.6 LTS Released
> Blender 3 taking so much time to give AMD Linux users hardware acceleration back makes me sad.

I believe the issue comes from AMD which disable/doesn't support their GPGPU libraries on Linux and Blender cannot do much about it. GPGPU support in general is much much better on NVIDIA thanks to CUDA, both for Windows and Linux.

artemisart··on Compiling Typed Python
I'm 1000% with you, I never understood why GAFAM never invested much in python despite using it quite heavily. Maybe more NIH and politics than engineering issues I guess?
artemisart··on Binance sees $2B in outflows as troubles compound
No, if you want to get your tokens out you're selling them to get something else than BNB (and BUSD) and this would drive the price down.
artemisart··on SQLAlchemy 2.0 Released
I find ORM models useful for type checking and migrations, what is your experience for this with pugsql?
artemisart··on I over-engineered a Fast Fourier Transform for Arduino
Yes but you have no choice for streaming/real-time applications or anytime you need results for each sample, where doing the full recompute would be O(n^2 log n) total.
artemisart··on JC – JSONifies the output of many CLI tools
How do you differentiate types with jevko (numbers, strings, boolean)? Your examples on jevko.org appear lossy as they encode in the same way things that are different in JSON and I don't know how you would then differentiate between true and "true", 27 and "27", etc.
artemisart··on JC – JSONifies the output of many CLI tools
I don't believe it's really an issue in practice, your `jc` alias will just take the priority and you can easily add one for jsonquery -> `path/to/bin/jc`. I think good short command names can help adoption (like for ripgrep, fd) but it's true that we should have a race to squat all the 2 letter names.
artemisart··on Show HN: TensorDock Core GPU Cloud – GPU servers from $0.29/hr
Yes... a full A100 (+ reasonable CPU etc) looks closer to 2€.
artemisart··on Ask HN: How to disable right-click blocking in the browser
I use F3 for this, I don't know if it's possible but I've never seen a website hijack F3.
artemisart··on SQLite Internals: Pages and B-trees
Embeddable... in the java ecosystem. I often see comments about datalog/datomic on HN and it seems interesting but I never see it anywhere else, is it because it's mostly known and used in the java and clojure ecosystem? Do you know of any free database with similar models usable from e.g. python?
artemisart··on Ask HN: What are some cool but obscure data structures you know about?
Cache oblivious data structures are absolutely used in every serious database.

From what I remember/can find online (don't take this as a reference): B-tree: relational DB, SQLite, Postgres, MySQL, MongoDB. B+-tree: LMDB, filesystems, maybe some relational DBs. LSM trees (Log-structured merge tree, very cool) for high write performance: LevelDB, Bigtable, RocksDB, Cassandra, ScyllaDB, TiKV/TiDB. B-epsilon tree: TokuDB, not sure if it exists anymore. COLA (Cache-oblivious lookahead array): I don't know where it's used.

Maybe modern dict implementations can qualify too, e.g. Python.

← PreviousPage 2 of 3Next →