Finally dipped my toes into RL post-training. I trained a code generation LLM with GRPO using open-r1. Here are my 9 takeaways: kevtimova.github.io/posts/grpo/
08.07.2025 18:29
π 2
π 0
π¬ 0
π 0
Finally dipped my toes into RL post-training. I trained a code generation LLM with GRPO using open-r1. Here are my 9 takeaways: kevtimova.github.io/posts/grpo/
I asked ChatGPT, Gemini, and Claude for a clever joke. They all gave me the same one. Either AI is merging into a hive mind⦠or humor has officially been solved mathematically!
The principle of least effort, from psychology, describes how we favor efficiency over effort. It aligns with System 1 (fast, intuitive) vs. System 2 (slow, deliberate) reasoning. AI faces a similar challenge: knowing when to rely on heuristics vs. deeper reasoning.
Happy Holidays and a Joyful New Year!
Wall sculpture series from βRandom Walks of Happinessβ, 2024
Celebrating Art: the Process, the Experiment, the Media.
#randomwalksofhappiness #happynewyear #happyholidays #abstractart #art #experimentalart #drawing #sculpture #wire #foundobjectsart #december
If you are into ML theory (RL or not) with a proven track record, and you are interested in an industry research position, PM me. Feel free to spread the word.
I gave a talk on Compositional World Models at NeurIPS last week π
The recording is now online: neurips.cc/virtual/2024... (for registered attendees; starts at 6:06:00)
Workshop: compositional-learning.github.io
Just 10 days after o1's public debut, weβre thrilled to unveil the open-source version of the technique behind its success: scaling test-time compute
By giving models more "time to think," Llama 1B outperforms Llama 8B in mathβbeating a model 8x its size. The full recipe is open-source!
Announcement from TMLR jmlr.org/tmlr/
"""
π£ Heads-up π£ that TMLR will pause new submissions over the upcoming holiday period from December 2 2024 to January 6 2025 (midnight AoE on both dates). We will resume accepting new submissions on January 7, 2025. Happy Holidays!
"""
An screenshot of UnicodeIt website
Write math on π¦ with UnicodeIt!
For example: ΞΈ β ββΏ or ppΜ
β ΞΌβΊΞΌβ»
Use website or install system-wide in Linux, macOS, or windows
www.unicodeit.net
(Created several years ago with @svenkreiss.bsky.social)
If you post a research paper with a picture of an animal, I will follow you. It's the law.
some little bluesky tips π¦
your blocks, likes, lists, and just about everything except chats are PUBLIC
you can pin custom feeds; i like quiet posters, best of follows, mutuals, mentions
if your chronological feed is overwhelming, you can make and pin make a personal list of "unmissable" people
A plot showing that reranking improves recall as we increase the number of reranked docs, but with increasing docs we diminishing returns and eventually a performance dip.
Mat is not on π¦βposting on his behalf!
It's time to revisit common assumptions in IR! Embeddings have improved drastically, but mainstream IR evals have stagnated since MSMARCO + BEIR.
We ask: on private or tricky IR tasks, are rerankers better? Surely, reranking many docs is best?
How many documents should you retrieve when using a reranker? The answer might surprise you!
Check out the excellent work from our intern Mathew on this important retrieval question. π
Google Scholar is twenty years (and one day) old today https://www.infodocket.com/2024/11/18/google-scholar-turns-20-20-things-you-didnt-know-about-scholar/
whiteboard with MUST: - 10k user signups - view post on web - app store - other pds federate - 3p labels/services etc
A whiteboard from our Jan 2023 team retreat where I wrote down our goals, and the top one was β10k user signups.β
Weβre growing by 10k users every 10-15 minutes right now.
I'm making a list of AI for Science researchers on bluesky β let me know if I missed you / if you'd like to join!
go.bsky.app/AcP9Lix
New here? Interested in AI/ML? Check out these great starter packs!
AI: go.bsky.app/SipA7it
RL: go.bsky.app/3WPHcHg
Women in AI: go.bsky.app/LaGDpqg
NLP: go.bsky.app/SngwGeS
AI and news: go.bsky.app/5sFqVNS
You can also search all starter packs here: blueskydirectory.com/starter-pack...
A starter pack for #NLP #NLProc researchers! π
go.bsky.app/SngwGeS
You can read about my research on emergent communication with adaptive compute at inference time and its connection to OpenAI's o1 model here: nyudatascience.medium.com/from-academi...
#ai #machinelearning #research
Hello Bluesky! π I'm an AI researcher with a PhD from NYU's Center for Data Science. My research focuses on representation learning for images and video, with an emphasis on self-supervised learning and regularization methods. Excited to connect and explore here! #ai #deeplearning #computervision
