Deyan Ginev
@dginev
Researching NLP for Math-rich documents. LaTeXML dev. Software Engineer at arXiv creator of http://ar5iv.org
This is the moment Lean officially "made it" to mainstream culture IMHO. Kudos for the great results @HarmonicMath !
This year, @HarmonicMath tested our advanced reasoning model Aristotle on the International Mathematics Olympiad (IMO), achieving Gold Medal-level performance and producing formally-verified proofs in Lean. We livestreamed our results from the HP Garage, check it out!
AMD teams contributing to the llama.cpp codebase. Great support from the community with the review process. Exciting to see this open-source collaboration!
github.com/ggml-org/llama…
BREAKING: we've partnered with @metaai and @paperswithcode to build a successor to Papers with Code (which was sunsetted yesterday) PWC, founded by @rbstojnic and @rosstaylor90 has been an invaluable resource for AI scientists and engineers over the years (and an inspiration…
Official results are in - Gemini achieved gold-medal level in the International Mathematical Olympiad! 🏆 An advanced version was able to solve 5 out of 6 problems. Incredible progress - huge congrats to @lmthang and the team! deepmind.google/discover/blog/…
Not Even Bronze: Evaluating LLMs on 2025 International Math Olympiad 🥉 matharena.ai/imo/ Nice blog post from the team behind MathArena: Evaluating LLMs on Uncontaminated Math Competitions (arxiv.org/abs/2505.23281) providing independent analysis of LLM performance on IMO.
😬
I've been using agents to run ML experiments for a while now, and it's all fun and games until the agent decides it doesn't like your evaluation method and decides to change it to get higher scores 😅
Ragged Blocks: Rendering Structured Text with Style by Sam Cohen arxiv.org/abs/2507.06460 very interesting method for rendering nested structure while respecting typographical layout
Happy to introduce Kimina-Prover-72B ! Reaching 92.2% on miniF2F using Test time RL. It can solve IMO problems using more than 500 lines of Lean 4 code ! Check our blog post here: huggingface.co/blog/AI-MO/kim… And play with our demo ! demo.projectnumina.ai
👀
As AI advances, our contribution is more and more original knowledge - meaning something that can’t be inferred from what exists digitally already by reasoning. Something like the result of an experiment. Maybe it should be written more natively for AIs instead of people, eg PDF…
🤝
Mathematical reasoning is at the heart of intelligence. Thrilled to partner with @tachim and @vladtenev and the @HarmonicMath team and lead their Series B as they build Mathematical Superintelligence: verified AI for high-stakes fields from mission critical software to finance…
New blog post: How AI is breaking traditional remuneration models technollama.co.uk/how-ai-is-brea…
We just released the best 3B model, 100% open-source, open dataset, architecture details, exact data mixtures and full training recipe including pre-training, mid-training, post-training, and synthetic data generation for everyone to train their own. Let's go open-source AI!
Introducing SmolLM3: a strong, smol reasoner! > SoTA 3B model > dual mode reasoning (think/no_think) > long context, up to 128k > multilingual: en, fr, es, de, it, pt > fully open source (data, code, recipes) huggingface.co/blog/smollm3
Great viral moment. (also, notice that you don't need review score hacking to get posted to a preprint server)
"in 2025 we will have flying cars" 😂😂😂
🗓️ The June 2025 arXiv articles are now in ar5iv. This month we passed 2.5 million processable sources (arXiv IDs) available via bulk download. Large latexml upgrades are getting finalized this summer - and a v0.9 release. A full update of ar5iv will follow.
🗓️The May 2025 arXiv articles are now in ar5iv.
Fun fact: We may be on track for 1 TB size in combined article sources for arXiv's 2025 submissions.
AI is so smart, why are its internals 'spaghetti'? We spoke with @kenneth0stanley and @akarshkumar0101 (MIT) about their new paper: Questioning Representational Optimism in Deep Learning: The Fractured Entangled Representation Hypothesis. Co-authors: @jeffclune @joelbot3000
I have been advocating tensor methods for almost decade and a half. Take a look at our tensor methods in deep learning from a few years ago arxiv.org/abs/2107.03436 Tensorly package allows defining tensor operations in Pytorch seamlessly tensorly.org @JeanKossaifi
what if attention operated in 3D? This paper introduces trilinear (2-simplicial) attention, and it might have just rewrite the current transformer scaling law by squeezing out the same accuracy with far fewer tokens.