paper of the day · 6 Nov 2026 · picked by Claude
Muyao Yuan, Liyan Xu, Weizhan Zhang et al. · 3 Nov 2026 · single gpu · code
LoCo shortens chain-of-thought by building DPO preference pairs at the exact points where a reasoning model starts overthinking, rather than rewarding shorter traces overall.
The pipeline is self-contained: sample math traces from a small open reasoning model, flag candidate overthinking states, freeze those prefixes, regenerate shorter continuations that still reach the verified answer, and train with LoRA DPO on the resulting state-conditioned pairs. This fits a single GPU if you use a 1.5B distilled model instead of the paper's 7B. The target is concrete: roughly 40% shorter completions on math benchmarks at equal or better accuracy, and you can check whether the savings carry over to code. Along the way you build a DPO training loop, a verifier-based data pipeline and a careful accuracy-versus-length evaluation, and you see how where a preference pair branches changes what the model learns.
41 hf upvotes · Byungwoo Jeon, Yoonwoo Jeong et al. · 30 Oct 2026 · discuss · multi-gpu · code · 43
emerging topic “Unified Multimodal Models”: 6 papers in 2 weeks vs 2 in the 18 days before · early traction: 8 upvotes/day in its first 5 days · from kaist-ai, whose recent “Look Only When Needed: Certificate-Guided Visual Routing for…” drew 115 upvotes/points · only 41 HF upvotes so far · official code
15 hf upvotes · Erkang Zhu, Yin Lin et al. · 2 Nov 2026 · discuss · 38
emerging topic “foundation model”: 6 papers in 2 weeks vs 4 in the 18 days before · early traction: 5 upvotes/day in its first 3 days · from alibabagroup, whose recent “Scaling Agentic Data for Long-Horizon Terminal Intelligence” drew 128 upvotes/points · only 15 HF upvotes so far
22 hf upvotes · Mingzhe Zheng, Yue Wu et al. · 4 Nov 2026 · discuss · code · 38
early traction: 11 upvotes/day in its first 2 days · from Tencent-Hunyuan, whose recent “Hy-Nomas: Towards Free-form Interactive World Generation” drew 112 upvotes/points · only 22 HF upvotes so far · official code
13 hf upvotes · Tianyu Liu, Jun Zhang et al. · 2 Nov 2026 · discuss · multi-gpu · code · 36
emerging topic “speculative decoding”: 7 papers in 2 weeks vs 7 in the 18 days before · early traction: 7 upvotes/day in its first 2 days · from Qwen, whose recent “VisionWeave: Weaving Elastic Visual Representations as a Nat…” drew 55 upvotes/points · only 13 HF upvotes so far · official code
26 hf upvotes · Qinuo Cao, Xin He et al. · 29 Oct 2026 · discuss · code · 36
early traction: 9 upvotes/day in its first 3 days · shares 2 authors with “Memadapter: Counterfactual Adaptation Against Memory-induced…” (48 upvotes/points) · only 26 HF upvotes so far · official code
39 hf upvotes · Guanting Dong, Zhichao Sun et al. · 3 Nov 2026 · discuss · cluster-scale · code · 35
early traction: 13 upvotes/day in its first 3 days · shares 22 authors with “CoworkOS:Scaling Agent Coworkers for Real-world Automation” (100 upvotes/points) · only 39 HF upvotes so far · official code
19 hf upvotes · Kunjun Li, Wenhao Chai et al. · 2 Nov 2026 · discuss · laptop / colab · code · 35
emerging topic “speculative decoding”: 7 papers in 2 weeks vs 7 in the 18 days before · early traction: 6 upvotes/day in its first 3 days · from UCBerkeley, whose recent “Correction Training for Diffusion Models” drew 42 upvotes/points · only 19 HF upvotes so far · official code
14 hf upvotes · Aryan Das, Vinay Kumar Verma et al. · 30 Oct 2026 · discuss · empirical analysis · multi-gpu · 35
emerging topic “Muon”: 3 papers in 2 weeks vs 1 in the 18 days before · early traction: 7 upvotes/day in its first 2 days · shares 2 authors with “SALD: Self-Referenced Advantage Learning for Diffusion Model…” (50 upvotes/points) · only 14 HF upvotes so far
18 hf upvotes · Zixian Gao, Penghao Luo et al. · 31 Oct 2026 · discuss · code · 35
early traction: 9 upvotes/day in its first 2 days · from AlayaLab, whose recent “OuroWorld: Bringing Any 3D World Alive as Diverse, Endlessly…” drew 44 upvotes/points · only 18 HF upvotes so far · official code
4 hf upvotes · Khai Le-Duc, Nguyen Kim Hai Bui et al. · 29 Oct 2026 · discuss · cluster-scale · code · 34
from StanfordUniversity, whose recent “Hybrid Reasoning Agents” drew 553 upvotes/points · only 4 HF upvotes so far · official code
Papers that haven't taken off yet but show signs they could: their concept is suddenly appearing in several new papers, they're gaining upvotes fast in their first days, their authors or lab recently had a well-received paper, and they're a runnable method with code. The more attention a paper already has, the more its score is discounted; the top 10% most-discussed are left out. Upvotes and comments here add a small boost (at most +0.1) that updates live. Data updated 6 Nov 2026, 00:00 UTC.