Trading Inference-Time Compute for Adversarial Robustness This paper explores how increasing inference-time compute improves the adversarial robustness of reasoning models (OpenAI o1-preview and o1-mini) without adversarial training. Problem: Adversarial attacks remain a major
@askalphaxiv
-
DeepRAG: Adaptive Retrieval for Large Language Models
By
–
DeepRAG: Thinking to Retrieval Step by Step for Large Language Models DeepRAG enhances retrieval-augmented generation (RAG) by modeling reasoning as a Markov Decision Process (MDP), enabling adaptive retrieval and improving answer accuracy. Problem: LLMs struggle with factual
-
AlphaGeometry2 Surpasses IMO Gold Medalists in Geometry
By
–
Gold-medalist Performance in Solving Olympiad Geometry with AlphaGeometry2 AlphaGeometry2 (AG2) surpasses IMO gold medalists in solving Olympiad geometry problems by expanding its problem domain, enhancing symbolic reasoning, and integrating a stronger language model. Problem:
-
Top arXiv Paper Trends: DeepSeek and Test-Time Scaling
By
–
Trends of top arXiv papers this past month Showing views on alphaXiv over time for
– DeepSeek V3
– DeepSeek R1
– s1: Simple test-time scaling -
alphaXiv: Alternative Platform for AI Research Papers Access
By
–
Check out http://
alphaXiv.org! Alternatively replace the “arXiv” with “alphaXiv” in any arXiv URL. Ex: https://
arxiv.org/abs/2501.17161 -> https://
alphaXiv.org/abs/2501.17161 -
Google AlphaGeometry 2 Achieves 84% on IMO Geometry Problems
By
–
New from Google: AlphaGeometry 2 Gets 84% over all IMO geometry problems from the last 25 years
-
How LLMs Learn to Reason: Comprehensive Study Reveals
By
–
Trending on alphaXiv (2/6): The most comprehensive (and refreshingly clear) study on how LLMs learn to reason. A systematic investigation reveals key ingredients: SFT initialization helps, reward shaping stabilizes training, filtered verifiable rewards improve generalization,
-
Cursor Uses Gemini 2 Flash for arXiv Paper Analysis
By
–
We used Gemini 2 Flash to build Cursor for arXiv papers
— alphaXiv (@askalphaxiv) 6 février 2025
Highlight any section of a paper to ask questions and “@” other papers to quickly add to context and compare results, benchmarks, etc. pic.twitter.com/2KKTuDuzRhWe used Gemini 2 Flash to build Cursor for arXiv papers Highlight any section of a paper to ask questions and “@” other papers to quickly add to context and compare results, benchmarks, etc.
-
alphaXiv Firefox Extension Launches for arXiv Paper Discovery
By
–
The alphaXiv Extension is now live on Firefox! Like papers directly on arXiv and see discussions on relevant papers in your community!
— alphaXiv (@askalphaxiv) 18 janvier 2025
Link to download: https://t.co/OcT79cM6Xq pic.twitter.com/nqdesvSn8IThe alphaXiv Extension is now live on Firefox! Like papers directly on arXiv and see discussions on relevant papers in your community! Link to download: https://
addons.mozilla.org/en-US/firefox/
addon/alphaxiv/
… -
Understanding Extrapolation: Causal Models for Out-of-Distribution Prediction
By
–
Towards Understanding Extrapolation: a Causal Lens The paper addresses extrapolation, where models predict outcomes for target samples outside the training distribution. It introduces a latent-variable model to formalize when extrapolation is feasible and offers methods to