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.
Post-Training Leaves Behavioral Shadows on Unrelated Decisions
Ziyang Zhang, Yubin Jing, Yuanhao Zeng, Yuyao Li et al.
Claims post-training skill transfers via single-word probes on unrelated prompts (+5.3pp HumanEval+); rigorous controls, but breadth results are underpowered.
AFT Neural Function Approximators for 1D Nonlinear Force Laws
Miriam Goldack, Johann Groß, Malte Krack, Merten Stender
SFNs replace AFT force evaluation in HBM; rigorously validated on 3 elements, but speedup demonstrated only for the hysteretic case and single-element systems.
Evaluating Communicative Success in Machine-Translated Conversation
Faiz Ghifari Haznitrama, Alice Oh
Checklist benchmark for interpreter-MT communicative success; judge validation is thorough, but model rankings carry an admitted context-asymmetry confound.
No More Free Lunch: Corpus Task Complexity Matters as Corpora Grow
Prasann Singhal, Amanda Bertsch, Jacob Steinhardt, Sewon Min
Defines Corpus Task Complexity and a 22-task benchmark showing sparse/hybrid attention degrades on O(N²) tasks; convincing, but fine-tuned, ≤32k evidence only.
Intrinsic Sequence-Likelihood Confidence in Retrieval-Dominated Extractive QA: Two Pre-Specified Negatives, and What They Do and Do Not Attribute
Gunwoo Lee, Changmin Sung, Sang-Hwan Gwak, Ina Kim et al.
Pre-specified negatives: intrinsic LLM confidence fails as distillation trigger and router in retrieval-dominated extractive QA; rigorous but setting-bound.
Policy as Code: A Coroutine-Bridge Harness for Fast-Reasoning Reliability on CAR-bench
Ivan Matveev
Competition report: a coroutine-bridge harness wins CAR-bench Track 2 at 60.0% Pass^3 via policy-as-code; n=30 hidden set limits statistical strength.
Baszta: Data-Centric Fine-Tuning of a Polish Multi-Label Safety Classifier
Adam Górski, Mateusz Jąkalak, Rafał Jakubowski
Reports a small but significant micro-F1 lead over Sójka on Polish safety classification; unusually honest self-critique, but OOD gains risk style overfitting.
Learnable Time-Frequency Masks for Explaining Time-Series Classifiers
Theresa Dahl Frehr, Francisco Pelayo, Lukas Raad, Alicia García Sanz et al.
XACT learns sparse attribution masks over STFT/wavelet coefficients for time-series classifiers; sound, honest, but code-free and evaluation-biased tuning.
BridgeMem: Causal Dyadic Transition Residuals for Temporal Knowledge Graph Forecasting
Zeyan Li, Libing Chen, Shengda Zhuo, Yin Tang et al.
Dyadic-transition residual atop frozen CENET wins all 20 metric comparisons on 5 TKG benchmarks; cheap and controlled, but tiny WIKI margins lack variance.
Generalization through Lexical Abstraction in Transformer Models: The Case of Functional Words
Giuseppe Samo, Vivi Nastase, Paola Merlo
Claims functional/lexical sentence variants share recoverable syntactic-semantic structure in Electra; suggestive VAE-probe results, but single-model, no stats.
Sufficiently Reduced Distributional Regression
Alexander Henzi, Tiange Liu, Xinwei Shen
SRDR recasts SDR as energy-score risk minimization with consistency guarantees and strong empirics; solid, but theory assumes exact empirical minimizers.
ALOE: Semantically Addressed Low-Rank Operators for Knowledge Editing
Zeyan Li, Hu Xu, Jianfeng Xu
Claims gated low-rank addresses fix edit scope (eff .955-.999, loc .981-1); strong mechanism work, but baselines look undertuned, generalization weak.
pylazaro: a Python package for anglicism extraction in Spanish
Elena Alvarez-Mellado
pylazaro unifies five Spanish anglicism-extraction models (F1 up to 0.86, well above LLMs) in one solid package; useful tool paper, but no new experiments.
PetriBench: Benchmarking LLM Reasoning over Dynamic State Spaces
Pyrros Koussios, Benjamin J\"ager, John Hua Yao, Ajay Sridhar et al.
Claims a scalable exactly-verified Petri-net LLM reasoning benchmark; credible design, but Hard-tier floor effects and single-run noise temper conclusions.
D-Quant: Driftable Entropy Coding for KV Cache Quantization
Yi Su, Hong Liu, Guanghua Yu, Jianchen Zhu
D-Quant fits entropy-coded KV into fixed per-token containers via drift, near-BF16 at 2.26 BPV; strong results, but only two 8B models, no code release.
Online Task Adaptation via Self-Organisation
Krsto Proroković
Meta-learned NCA fast memory gives gradient-free one-pass CIFAR-100 adaptation (20→48.2% vs 54.4% scratch); method transparent, eval scope small.
Evolution or Illusion? Rethinking Evaluation in LLM Evolutionary Search
Tal Oved, Roi Pony, Oshri Naparstek, Udi Barzelay
Shows single-point budgets can misrank LLM evolutionary search; 40-seed×200-iter replay grid makes the case, though scope is narrow.
From Text Decisions to Pixels: An Study of Jev-Style Visual Choice Model
Xunlan Zhou, Xianliang Yang, Li Zhao
PixelJev is standard candidate-logit VLM readout + LoRA framed as a decision interface; rigorous, honest evaluation, thin novelty, no Visual Jev comparison.
JustMem: Just-Enough Memory Access for Long-Term Conversations
Guanhua Chen, Yanting Wang, Wenjing Zhi, Lei Sha
Proposes query-adaptive memory access (LOOKUP/COMPOSE/REPLAY); strong efficiency-accuracy results, but LoCoMo edge over LightMem is not significant (p=0.53).
Zarya: A Hybrid Autoregressive--Masked Diffusion Language Model with Flexible Training and Dual-Mode Inference
Leonid Sinev, Ilya Koziev, Vladislav Leshchuk
Hybrid AR+MDM training with slot curriculum and dual-mode KV-cached inference; sound design, but evaluation covers one model with no baselines or speed data.
F$^{2}$DR: A Fine-Grained Full-Pipeline Reward Framework for DeepSearch Workflows
Bojian Xiong (Tianjin University), Wentao Ding (Baidu Inc.), Yujing Lu (Baidu Inc.), Shaowei Zhang (Tianjin University) et al.
Promising three-dimension process-level reward framework + 1,070-pair DeepSearch RM benchmark; strong gains, but circularity and verbosity confounds remain.
Dictionary-Constrained Grapheme-to-Phoneme for Unsegmented Languages from LLM-Annotated Data
Rui Hu, Zhenpeng Zhan, Xiaolong Lin
Lattice-CRF Japanese G2P trained on 2.3M LLM sentences hits 99.62% accuracy, beating all neural baselines; convincing, though frontier LLMs still edge it out.
From Parameters to Behaviors: A Survey of Model Fusion for Large Language Models
Shuo Cai, Yanggan Gu, Zihao Wang, Yuanyi Wang et al.
Claims first unified 3-level taxonomy of LLM model fusion; clear definition and honest comparisons, but taxonomy boundaries and selection method are debatable.
A Phonemically Comprehensive, ASCII-Only Romanization Scheme for Thai and Lao: Systematic Cross-Lingual Correspondence and Chinese-User-Friendly Design
Zijie Zhang, Tan Lee
A fully specified ASCII romanization unifying Thai and Lao phonology with Pinyin/Jyutping-aligned symbols; coherent design, but entirely unevaluated.