unsupervised learning and rl
San Francisco, CA
Joined April 2008
- Tweets16.2K
- Following2.5K
- Followers2.7K
- Likes19.6K
Volodymyr Kyrylov retweeted
An internal model at OpenAI has produced a claimed refutation of ETH! Russell Impagliazzo, Mohan Paturi, and I have been asked to help organize an open and independent examination of this claim. 1/4
Volodymyr Kyrylov retweeted
If you've wondered where I've been recently, I gave an interview. This is a very difficult time for me. I'm trying to bring my mom home, and I could use your help spreading the word. Please watch, and please share.
interview: youtube.com/watch?v=JlK2a6yv…
website: bringmymomhome.com/
Volodymyr Kyrylov retweeted
The core of ai psychosis, by the way, is the failure to properly align human-intelligence axes with ai-intelligence ones.
Volodymyr Kyrylov retweeted
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.
Volodymyr Kyrylov retweeted
as of 5 minutes ago, our gpt-oss implementation is merged into torchtitan!
thanks to all the work by @jianiw_wang @__tianyu at @pytorch making it clean & scalable for the community ❤️
i hope y'all play around with training gpt-oss, it's great for its sparsity & reasoning
periodic ❤️ open-source!
for example, we’ve been collaborating with the @PyTorch team to build the highest MFU gpt-oss training implementation (includes thinky sinky flexattn)
here’s a few SFT runs of gpt-oss-20b & 120b, where i get ~24% MFU for 20b and ~8% for 120b
Volodymyr Kyrylov retweeted
LLMs are winning IOI golds & crushing code gen—but can they verify correctness?
In Feb, our benchmark saw single digit scores with o3-mini. We re-ran our evals with the latest open models: GPT-OSS gets 21.6% at demonstrating bugs in code! Progress✅But verification's still hard
Volodymyr Kyrylov retweeted
👀 we care a lot about correctness, ran many evals and stared at many tensors to compare them. numerics of vLLM on hopper should be solid and verified! if you run into any correctness issue on vLLM, we would love to know and debug them!
Volodymyr Kyrylov retweeted
Yes, I’m very sure vllm is correct — we spent quite a bit of time on that. 🥹
Volodymyr Kyrylov retweeted
Very excited to see this model released to the open-source community. It's still hard to believe that our latest techniques allow it to be so incredibly powerful yet so remarkably small.
super excited to have contributed to gpt-oss. We have put a lot of love into both training the model and making the developer examples, check them out: github.com/openai/gpt-oss
Volodymyr Kyrylov retweeted
Today we release gpt-oss-120b and gpt-oss-20b—two open-weight LLMs that deliver strong performance and agentic tool use.
Before release, we ran a first of its kind safety analysis where we fine-tuned the models to intentionally maximize their bio and cyber capabilities 🧵
Volodymyr Kyrylov retweeted
"i don't fear computers, i fear the lack of them" is a quote form Isaac Asimov's Foundation series, prompted into this little animation, which started as an attempt to reproduce ASCII binary flow but somehow got transformed into this. anyway, its cool chatgpt.com/canvas/shared/68…
Volodymyr Kyrylov retweeted
great people at the ai meetup today!
also: dolores park is really beautiful
with @arankomatsuzaki and @iScienceLuvr
Volodymyr Kyrylov retweeted
Introduction to deep research
東京からのライブ🗼
openai.com/live/
Volodymyr Kyrylov retweeted
o3-mini is here! Together with @shengjia_zhao, @_kevinlu, @max_a_schwarzer, @ericmitchellai, @brian_zq, @sandersted and many others, we trained this efficient reasoning model, maximally compressing the intelligence from big brothers o1 / o3. The model is very good in hard math/coding/science questions with a fraction of cost and latency, defining new cost-efficient reasoning frontier.
The model supports three different reasoning powers. Users can adjust the thinking time based on different use cases. The longer the model thinks, the better its capability is.
With o3-mini-low, we drastically reduce latency compared to o1-mini, achieving GPT-4o level latency for response.
You can apply early access to this model today to do safety-testing!
openai.com/index/early-acces…
Volodymyr Kyrylov retweeted
@miniapeur here's a challenge for you: can you do a chad-meets-chad meme where Kolmogorov says "universal compression is impossible" and Shannon says "universal compression is possible" and they shake hands in mutual respect?
Volodymyr Kyrylov retweeted
Wrote a post about Highway networks, ResNets and subtleties of architecture comparisons
Volodymyr Kyrylov retweeted
[LG] Key-value memory in the brain
S J. Gershman, I Fiete, K Irie [Harvard University & MIT] (2025)
arxiv.org/abs/2501.02950
Volodymyr Kyrylov retweeted
Highly recommend learning by doing, but if you want an intro into audio DSP I recommend starting somewhere easier, e.g., a resampler, a noise suppressor, an AGC. Writing a software AEC (with delay estimation and NLP) as a first project is something out of Good Will Hunting.
