Spurious Rewards Paradox: Mechanistically Understanding How RLVR Activates Memorization Shortcuts in LLMs
https://arxiv.org/abs/2601.11061
via @buckwheat_thoughts
https://arxiv.org/abs/2601.11061
via @buckwheat_thoughts
arXiv.org
Spurious Rewards Paradox: Mechanistically Understanding How RLVR...
Reinforcement Learning with Verifiable Rewards (RLVR) is highly effective for enhancing LLM reasoning, yet recent evidence shows models like Qwen 2.5 achieve significant gains even with spurious...
Weak Diffusion Priors Can Still Achieve Strong Inverse-Problem Performance https://arxiv.org/abs/2601.22443
arXiv.org
Weak Diffusion Priors Can Still Achieve Strong Inverse-Problem Performance
Can a diffusion model trained on bedrooms recover human faces? Diffusion models are widely used as priors for inverse problems, but standard approaches usually assume a high-fidelity model trained...
Are AI Capabilities Increasing Exponentially? A Competing Hypothesis https://arxiv.org/abs/2602.04836
arXiv.org
Are AI Capabilities Increasing Exponentially? A Competing Hypothesis
Rapidly increasing AI capabilities have substantial real-world consequences, ranging from AI safety concerns to labor market consequences. The Model Evaluation & Threat Research (METR) report...
BabyVision: Visual Reasoning Beyond Language https://unipat.ai/blog/BabyVision
UniPat AI
BabyVision: Visual Reasoning Beyond Language
State-of-the-art MLLMs achieve PhD-level language reasoning but struggle with visual tasks that 3-year-olds solve effortlessly. We introduce BabyVision, a benchmark revealing the infancy of AI vision.
Forwarded from Hacker News
The Waymo World Model: A New Frontier for Autonomous Driving Simulation (π₯ Score: 157+ in 1 hour)
Link: https://readhacker.news/s/6Ma63
Comments: https://readhacker.news/c/6Ma63
Link: https://readhacker.news/s/6Ma63
Comments: https://readhacker.news/c/6Ma63
Waymo
The Waymo World Model: A New Frontier For Autonomous Driving Simulation
We are excited to introduce the Waymo World Model, a frontier generative model that sets a new bar for large-scale, hyper-realistic autonomous driving simulation.
Refer-Agent: A Collaborative Multi-Agent System with Reasoning and Reflection for Referring Video Object Segmentation https://arxiv.org/abs/2602.03595
arXiv.org
Refer-Agent: A Collaborative Multi-Agent System with Reasoning and...
Referring Video Object Segmentation (RVOS) aims to segment objects in videos based on textual queries. Current methods mainly rely on large-scale supervised fine-tuning (SFT) of Multi-modal Large...
Learning to Repair Lean Proofs from Compiler Feedback https://arxiv.org/abs/2602.02990
arXiv.org
Learning to Repair Lean Proofs from Compiler Feedback
As neural theorem provers become increasingly agentic, the ability to interpret and act on compiler feedback is critical. However, existing Lean datasets consist almost exclusively of correct...
β€1
Universal Topological Gates from Braiding and Fusing Anyons on Quantum Hardware https://arxiv.org/abs/2601.20956
arXiv.org
Universal Topological Gates from Braiding and Fusing Anyons on...
Topological quantum computation encodes quantum information in the internal fusion space of non-Abelian anyonic quasiparticles, whose braiding implements logical gates. This goes beyond Abelian...
Expanding the Capabilities of Reinforcement Learning via Text Feedback https://arxiv.org/abs/2602.02482
arXiv.org
Expanding the Capabilities of Reinforcement Learning via Text Feedback
The success of RL for LLM post-training stems from an unreasonably uninformative source: a single bit of information per rollout as binary reward or preference label. At the other extreme,...
Goal-Guided Efficient Exploration via Large Language Model in Reinforcement Learning https://arxiv.org/abs/2509.22008
arXiv.org
Goal-Guided Efficient Exploration via Large Language Model in...
Real-world decision-making tasks typically occur in complex and open environments, posing significant challenges to reinforcement learning (RL) agents' exploration efficiency and long-horizon...
Self-dual Higgs transitions: Toric code and beyond https://arxiv.org/abs/2601.20945
arXiv.org
Self-dual Higgs transitions: Toric code and beyond
The toric code, when deformed in a way that preserves the self-duality $\mathbb{Z}_2$ symmetry exchanging the electric and magnetic excitations, admits a transition to a topologically trivial...
π€―2πΎ1
We hid backdoors in binaries β Opus 4.6 found 49% of them https://quesma.com/blog/introducing-binaryaudit/
Quesma
We hid backdoors in binaries β Opus 4.6 found 49% of them - Quesma Blog
BinaryAudit benchmarks AI agents using Ghidra to find backdoors in compiled binaries of real open-source servers, proxies, and network infrastructure.
π6
An X-ray-emitting protocluster at z β 5.7 reveals rapid structure growth https://www.nature.com/articles/s41586-025-09973-1
Nature
An X-ray-emitting protocluster at z β 5.7 reveals rapid structure growth
Nature - Discovery of a protocluster at z = 5.68, merely one billion years after the Big Bang, suggests that large-scale structure must have formed more rapidly in some regions of the...
Soft Contamination Means Benchmarks Test Shallow Generalization https://arxiv.org/abs/2602.12413
arXiv.org
Soft Contamination Means Benchmarks Test Shallow Generalization
If LLM training data is polluted with benchmark test data, then benchmark performance gives biased estimates of out-of-distribution (OOD) generalization. Typical decontamination filters use n-gram...