Evolve Quick Reviews
Public Paper Reviews
Structured preliminary reviews of recent open-access papers, generated and continuously updated by the Evolve analysis engine. Each review condenses what the paper claims, how solid the evidence is, and the questions a rigorous peer reviewer would ask — English analysis with abstracts in your language. These are independent AI analyses, not peer review; the canonical record always lives at the publisher.
Small Cues, Big Consequences: Learning Pivotal Cues for Multimodal Meme Classification
Akshit Sharma, Prashant W. Patil
MemeCF (LLM-annotated cues) + MemePIVOT (UOT grounding, evidential fusion) beat strong baselines on 3 meme benchmarks; LLM-derived labels merit scrutiny.
Diffusion-Induced Spatial Attention Overlapping Community Detection
Kosti Koistinen, Vesa Kuikka, Joni Herttuainen, Matthew Hendren et al.
DISCO biases sparse attention with an ISM diffusion prior for overlapping communities; competitive on six small Facebook ego-nets, but narrow scope, no code.
Optimal Sequential Annotations for Off-Policy Evaluation
Woojin Chae, Ezinne Nwankwo, Haitong Qin, Angela Zhou
Variance-optimal annotation for sequential OPE: sound theory, strong empirical gains (up to 68% RMSE cut), but real-data gains benchmarked vs silver labels.
SpeakerMem-R1: Speaker-Centered Dual-Track Memory for Multi-Party Dialogue
Haobo Zheng, Tan Tang, Yan Chen, Weijie Wang et al.
Dual-track speaker-attributed memory + RL-trained 3B writer shows solid multi-party gains; thin RL data, config selection and weak LoCoMo transfer temper it.
Measuring the Serving Stack Instead of the Model: Hidden Confounds in Local Tool-Use Evaluation
Lijuan Tang, Yuemeng Zheng
A measurement study showing Ollama's per-model tools gate and lossy harness logging can report capable local models at 0% fidelity; honest but thin-n evidence.
From Alignment to Access Control: A Framework for GenAI Policy Enforcement
Nathalie Baracaldo
Position paper unifying GenAI policy meanings; argues prompt-based 'Pretty Please' enforcement is unsafe, hard controls needed. Persuasive but unvalidated.
Beyond Repeated Sampling: Learning Search Policies for LLM Reasoning
Ismail Labiad, Matthieu Kowalski, Marc Schoenauer, Rémi Munos et al.
RL-trains a small concept generator to steer a frozen larger LLM, ~doubling pass@128 on hard math; strong controls, but judge bias and training cost remain.
When are bosonic Gaussian states classical to learn?
Senrui Chen, Antonio Anna Mele, Francesco Anna Mele, John Preskill
Tight crossover result: cold bosonic Gaussian states need Ω(n³) copies (vs classical Θ(n²)); warm states hit Θ(n²) via simple heterodyne. Rigorous, complete.
Train Where the Quantized Model Goes: On-Policy Distillation for Low-Bit Reasoning
Yuanteng Chen, Zhilei Liu, Peisong Wang, Yuantian Shao et al.
On-policy distillation after QAD restores sub-3-bit reasoning (MATH retention 35%→70%); well-controlled but single-run evidence and partial ablations.
Metrics Failure in LLM-Based Code Vulnerability Repair: An Empirical Study and a Change-Aware Screen
Om Nepal, Sushant Aryal, Oluseyi Olukola, Nick Rahimi
Five controlled experiments convincingly show compile rate misleads in LLM vulnerability repair; valid for small models; diff_F1 screen unvalidated.
The Sirens' Song: When Proximal Background Context Overshadows Distant Evidence
Xiaoyu Yang, Jie Lu, Wei Duan, En Yu
Claims the 'Proximity Trap' and LYRA, a t-distributed QK rescoring variant with gains on LongBench-v2/RULER/LongBench; promising diagnosis, but gains may partly stem from extra fine-tuning, and single-run stats are unverified
CORE-STACK+: Meta-Learning for Deep Stacked Generalization
Noor Islam S. Mohammad
Stacking-method paper with strong claimed gains, but internal number clashes, an OOF leakage contradiction, and a flawed ECE proof undermine trust.
TraceVIC: Causal Reasoning over Code Evolution for Identifying Vulnerability-Inducing Commits
Fnu Tanish, Samiha Shimmi, Samikshya Chapagain, Hamed Okhravi et al.
Claims evolution-aware temporal-graph ranking beats SZZ for VIC identification (F2 0.820 vs 0.666); strong evaluation but no uncertainty estimates.
Type-Safe Is Not Error-Free: A Constrained Decision Head Follows the Option Name, Not the Rubric Bound to It
Yu Sun, Junhao Xu, Jiajia Shi, Zijin Yang
Claims typed models follow option names over definitions (Laya AUC 93.8→23.2% on yes/no swap); well-controlled, but only two encoders plus a closed API model.
Rolling Conformal Prediction in Sequential Model Training
Chen Cheng, Ruiting Liang, Rina Foygel Barber
Proposes rolling-CP: calibrate-then-train on streams with 1−2α coverage, no stability. Elegant tight theory; factor-2 gap and exchangeability limit practice.
Automatic depth-based local center clustering via $β$-integrated local depth and adaptive grouping
Siyi Wang, Alexandre Leblanc, Paul D. McNicholas
A-DLCC claims fully parameter-free depth-based clustering and largely delivers, but hidden fixed constants, no sensitivity analysis, and a Seg failure temper the claim.
An Exposition of GPT Astra's Proof of Lower Bound on DP Continual Counting
Jalaj Upadhyay
严谨的阐述验证了HL26关于GPT Astra的最优纯DP连续计数下界的论述；证明基本无误，但未提出新的定理。
Does AI Save Time on Product Design? A Randomized Controlled Experiment of AI Prompt-to-Design Workflows
Remy Stewart, Olabode Anise, Andrew Hogan, Augustus Griffin
A genuinely credible RCT with good statistics, but completer-only time analysis, garbled tables, and no COI statement (all authors are Figma employees) temper the 20% claim.
Agensh: Scaling Organizational Intelligence to 1,024 Agents
Zhihao Zhan, Ting Song, Li Dong, Shaohan Huang et al.
Reports orchestrator-free harness scaling to 1,024 LLM agents; +49% rel. gain at 128 agents on ProgramBench, but no variance, cost or orchestrator baseline.
EquivSVA: A Formally Verified Dataset of Behavioral Assertions Across Equivalent RTL Implementations
FNU Aditi
120-family formally verified RTL assertion dataset; rigorous validation, reproducible artifacts; synthetic origins and single-model demo limits validity.
Grow the Harness, Not the Context: From Strategy-Free Scaffolds to Reusable Specialist Agents
Laizhen Li, Jiarui Li, Juanjuan Zhao, Kejiang Ye et al.
Learns agent harness code from failures starting from an empty scaffold; strong cost–success gains but tiny eval sets and no learned-baseline comparison.
Which Objectives Need a Dial? Predicting Objective Conflict and Covering Trade-offs in Steerable Pluralistic Alignment
David Tsoi, Esra Dönmez
Cheap pre-training metrics predict MODPO objective conflict on HelpSteer but fail on AI-annotated UltraFeedback; honest work, single-seed, seven pairs.
Harness as a Language: A Minimalist Agent Framework With Maximal Expressivity
Zhening Li, Joshua Liu, Mateja Vukelic, Nicole Shen et al.
JAZ's 'invoke' loop (all inputs + history as REPL variables) rivals specialized memory/self-improvement harnesses on 2 benchmarks at lower cost, but margins are thin and untested statistically.
CliffCompaction: Cost-Efficient Compaction for Long-Horizon Coding Agents
Trang Nguyen, Eulrang Cho, Bingqing Chen, Tim Dettmers
Truncation-only compaction cuts coding-agent cost ~50% at maintained performance with strong KernelBench results; stats testing and fair baselines are thin.