Sabitlenmiş Tweet
Happy to release Accelerated Scan, a kernel library for first order parallel associative scans in vanilla @PyTorch, Triton 2.2.0 and CUDA C++.
pip install accelerated-scan🧵
English
Volodymyr Kyrylov
16.2K posts
Volodymyr Kyrylov
@darkproger
unsupervised learning and rl
San Francisco, CA Katılım Nisan 2008
2.4K Takip Edilen2.7K Takipçiler
highly encourage folks to read the gpt-4.5 blog. it includes lots of beautiful charts and graphs and demos and words. and 35 evals. openai.com/index/introduc…
English
This is a painful moment. We wanted him on the start line. The Olympic Games must remain a place where athletes compete together under the same rules.
IOC President Kirsty Coventry addresses the decision preventing skeleton pilot Vladyslav Heraskevych from starting his race at Milano Cortina 2026.
📰 Full story 🔗: oly.ch/4rgUolv
English
spark is close to an order of magnitude latency improvement for token output, now tools are the bottleneck. great for networks of codex exec
OpenAI@OpenAI
GPT-5.3-Codex-Spark is now in research preview. You can just build things—faster.
English
During my first attempts to exit vim, I didn’t realize I was accidentally entering codex from the 70s
English
Training on intermediate solutions during search helps kernel and algorithm engineering — excited about this application of gpt-oss.
Mert Yuksekgonul@mertyuksekgonul
How to get AI to make discoveries on open scientific problems? Most methods just improve the prompt with more attempts. But the AI itself doesn't improve. With test-time training, AI can continue to learn on the problem it’s trying to solve: test-time-training.github.io/discover.pdf
English
@dmsobol @jahulas @maciejpioro @kuba_krj @Alibaba_Qwen @BytedanceTalk @_selebou @SkyLi0n @ChengZhoujun @QuentinAnthon15 @xidulu @aman_gif We didn’t publish any measurements however I found a very detailed write up from @SupritiVijay @AmanPriyanshu6 that might be useful: github.com/AmanPriyanshu/…
English
@jahulas @maciejpioro @kuba_krj @Alibaba_Qwen @BytedanceTalk @_selebou @SkyLi0n @ChengZhoujun @QuentinAnthon15 @xidulu @aman_gif @darkproger how did gpt oss look like in terms of expert specialisation? I still haven’t had a chance to look at its weights and measure it.
English
MoE models are compute efficient. Everyone knows that. But they are not parameter efficient. Why?
Our experts learn redundant, overlapping functions. We're spending extra money loadings weights for redundant experts. This is wasteful. And we're not fixing it.
Attaching here a plot I showed at my @NeurIPSConf workshop showing how much redundancy there is in deepseek-v3 model in the first 23 layers. Diagonal is expert compare to itself. 🧵
English
@jachiam0 @IrvZinter Save as mp4, ffmpeg has ok default visualizations
English
@IrvZinter I did and I didn't find the answers satisfying. Google Drive? Dropbox? Sort of ridiculous, this feels like the shape of problem that should have been solved via a specific and easy optimized garden path
English
I'm actually a little shocked it took me this long to realize, but I can get ChatGPT to make music by having it write code to generate MIDI files. Time to see if it can compose RPG battle music that is any good
English
macapptree is an amazing tool for screen perception in gpt-oss!
Mariya Hirna 🇺🇦@HirnaMariya
Wild NeurIPS moment: @darkproger from @OpenAI told me he uses our open-source macapptree as his go-to tool for parsing macOS accessibility 🤯 Made my day!
English
lol Neurips - 2600+ people registered for our bar crawl and after party 🤯
English
Volodymyr Kyrylov retweetledi
The Strategic Explorations team @OpenAI is looking to recruit researchers interested in working on the next frontier of language modeling!
Feel free to reach out to me by email. @darkproger and I will also be at NeurIPS to connect and discuss in person.
English
@Miles_Brundage gpt-oss has all kinds of tables for you in store
English
