Scaling Laws & Emergent Abilities - Weekly Roundup

This survey systematizes the EML landscape by introducing the first structured, model-to-system taxonomy and establishes a structured framework for multimodal systems that are not only high-performing and generalizable but natively efficient and ready for ubiquitous deployment.

Probe of Internal Recognition (PIR) separates a model that will not answer from one that cannot, which supports sandbagging audits and unlearning verification, and extends from multiple-choice questions to free-form generation.

ALIBI is presented, a semantic cover story attack against frontier LLM-based malware analyzers that adds a small, non-executed read-only section to a compiled binary, containing a coherent but false security product narrative, without altering imports or executable behavior.

This study introduces TrustReviewer, an open-source LLM-based system for generating peer reviews of AI and machine learning papers and describes a concrete risk of recursive reviewer training and provides practical interventions for preserving judgment diversity and improving recommendation alignment in AI-assisted scientific evaluation.

Probe of Internal Recognition (PIR) separates a model that will not answer from one that cannot, which supports sandbagging audits and unlearning verification, and extends from multiple-choice questions to free-form generation.

ALIBI is presented, a semantic cover story attack against frontier LLM-based malware analyzers that adds a small, non-executed read-only section to a compiled binary, containing a coherent but false security product narrative, without altering imports or executable behavior.

This study introduces TrustReviewer, an open-source LLM-based system for generating peer reviews of AI and machine learning papers and describes a concrete risk of recursive reviewer training and provides practical interventions for preserving judgment diversity and improving recommendation alignment in AI-assisted scientific evaluation.

REARL, a closed-loop simulation enhancement framework that integrates real traffic data with large language model (LLM) and modulates the simulated vehicle with reference to that real action, is proposed.

A 202-scenario benchmark with Reference Decisions under a seven-class taxonomy is introduced and Decision Alignment and safety-specific error metrics are evaluated, finding structured LLM decisions are insufficient as a standalone safety mechanism and deployment requires an independent enforcement layer.

MERIT-Rank is proposed, a framework that models complementary reasoning trajectories to improve reranking robustness and develops Progressive Rank Policy Optimization (PRPO), a progressive training framework that stabilizes reasoning trajectories while continually improving ranking quality through staged optimization objectives.

This work instantiates VDN on MiniMax H3, which combines local Softmax attention with bidirectional linear memory for long-range video context and introduces Video Delta Attention (VDA), which updates memory once per frame by jointly incorporating its spatial tokens.

It is shown that the instability of RL under TIM is primarily caused by drift: a persistent bias between training and inference engines that accumulates with every training step, and an additive score centering term is derived that stabilizes RL under TIM by canceling drift.

This work proposes MATCH, a closed-loop framework for model-aware tool learning with curriculum scheduling and hierarchically gated rewards, and shows consistent improvements across four backbones from two model families.

The aim of interpretability shifts the aim of interpretability from finding human concepts inside models toward discovering and characterizing the representational structures that are native to the models themselves and might affect their behaviour in unpredictable ways.

This work proposes a framework for locally masked SAE-based anomaly detection, supported by theoretical justifications, and demonstrates their ability to capture meaningful safety information while using only 1-2% of SAE neurons for computation.

Inspired by recent findings that self-attention acts as a low-pass filter and that its low-frequency components are informative, AttnPrint for white-box provenance is developed, which achieves strong derivative-model detection performance while remaining robust to five downstream modification techniques.

This tutorial aims to walk the reader through the journey from words to network traffic, shedding light on how language is translated into communication flows within HPC training systems, and tries to demystify how words travel across the network.

Decoupled-Live-ProsodyJudge (D-LPJ), a cost-effective pairwise evaluator distilled from Gemini into Qwen3-Omni, and a novel span-local GRPO strategy that applies normalized advantages strictly to their corresponding rationale spans.

KoNeoBench is a benchmark for evaluating LLMs' understanding of Korean neologisms, built on 1,785 Korean neologisms attested in online news since 2020 and curated through expert lexicographic review.

A Retrieval-Augmented Generation framework for scientific image quality assessment, designed to simultaneously address both the understanding track (SIQA-U) and the scoring track (SIQA-S) of the SIQA challenge, is proposed.

V\={a}kQA, a Telugu SQA benchmark of 2,001 factoid question-answer pairs across six domains, is introduced and it is observed that Telugu phrasing retains cultural specificity that is lost in translation, speech input introduces phonetic confusions that alter question meaning, and cascaded ASR-MT errors compound progressively.

The Profiling, Investigation, and Judgment (PIJ), comprising 2,500 real homicide cases from five countries, is introduced and it is found that performance degrades systematically as tasks shift from explicit fact extraction to implicit reasoning over unknown suspect profiles.

This work proposes a practical framework for verifiable DP training using CPU-side TEEs together with untrusted GPUs, and addresses a fundamental efficiency-security tension: training entirely inside a CPU TEE is too slow, while unrestricted GPU offloading can allow malicious deviations from DP.

Inspired by the human annotation training process, ProKDA first uses an agentic background knowledge construction pipeline to obtain external knowledge related to meme understanding, and adopts a three-stage training strategy that sequentially performs background knowledge learning, hatefulness detection learning, and hatefulness boundary alignment.

This survey defines model fusion and organizes prior work into three levels: parameter-level, representation-level, and behavior-level fusion, which provides a clear map of this area and support future work on model fusion.

EviRCA is presented, a framework for LLM-based RCA that decouples deterministic evidence extraction from LLM reasoning and substantially outperforms prior OpenRCA baselines that achieve up to 15.2%, while reducing token consumption by 15-26x and execution time by 3-20x.

Trivet is presented, a framework combining large language models (LLMs) and Lean for automated translation validation of LLVM transformations, which generates structured proof scaffolds based on source and target functions, automatically discharges obligations amenable to deterministic reasoning, and delegates transformationspecific obligations to LLMs.

AgriGround is introduced, a large-scale pixel-grounded agricultural multimodal instruction-tuning dataset containing over 500K images and 11M instruction-following samples spanning plant disease analysis, crop and weed identification, insect pest recognition, and fine-grained botanical understanding.

QCPruner fuses two cross-modal cues into utility and applies it to both visual targets and candidate representatives within visual-affinity-based coverage, which achieves the highest average relative performance among evaluated complete-system pruning methods at every reported token budget.

Xronos is proposed, a collaborative fine-tuning framework for heterogeneous CPU edge devices that uses TP as its execution backbone and combines lightweight profiling with heterogeneity-aware tensor partitioning to reduce the straggler bottleneck.

This work tests a model's ability to rank conformers by stability, suggesting that it emerged unintendedly from models training on linked but conceptually different tasks, and suggests that current frontier models offer a realistic starting point for AI-driven molecular design.

CoFree (Collapse-Free Reasoning Embedding), a two-stage framework that progressively integrates LLM reasoning into query and document embedding optimization while preserving reasoning quality, is proposed.

A fine-grained comparison of the capacity for parallelism among these three leading approaches is initiated and it is proved that uniform and Gaussian diffusion can sample in a number of forward passes which scales with the dual total correlation of the underlying distribution.

Adaptive Uncertainty-Routed Analysis is introduced, a multimodal email threat detection system that analyzes both the content of an email and its embedded URLs and quantifies prediction uncertainty from a URL classifier.

A decomposed confidence layer along three interpretable channels, including perception, layout, and validation is introduced, together with a final conformal risk control, which can be used for reliable STP of financial documents.

SALR is proposed, a schema-anchored latent reasoning method for LF construction that performs multi-step reasoning by generating continuous thoughts in the model's hidden states, thereby delaying the explicit commitment to LF decisions.

F2DR evaluates DeepSearch workflows across three dimensions: Content, Trajectory, and Answer, enabling comprehensive process-level assessment, and constructs DeepSearch RM-Bench, a dedicated benchmark for evaluating RMs in DeepSearch scenarios.

These findings show that robust multilingual healthcare evaluation requires explicit testing of register variation, uncertainty handling, and the provenance and invariance of adaptation labels.

It is shown that the language of a question can change how the same AI systems assess the war in Ukraine, suggesting a possible route through which information warfare may shape the text used to train AI models, which may in turn spread geopolitical biases.

The findings of MaSCoD support structural pre-organization as an explicit design and evaluation target for omission control and motivate evaluating context construction jointly with its utilization in judgment.

The results argue that useful self-reports remain sensitive to elicitation, correlated errors, and benchmark noise, and that useful self-reports remain sensitive to elicitation, correlated errors, and benchmark noise.

A Physics Informed Recurrent Neural Network (PIRNN) which predicts, along the target, unobservable variables on both historic data and forecast target, which enhances the model robustness and results interpretation using domain knowledge.

This work defines the task of safety as a highly pragmatic and critical cognitive dimension, and proposes the Cognitive Expected Scene Graph (CESG) Score, a structure-grounded, polarity-aware evaluation metric.

This work proposes Abstract Token Curriculum (ATC), a novel curriculum learning framework that elicits effective continuous intermediate representations without direct supervision or manual scratchpad design and shows ATC's effectiveness on graph reachability and arithmetic learning tasks.

This paper presents a context-aware neural G2P method that scores paths of a discriminative conditional random field over a word lattice constructed from dictionaries that strongly outperforms conventional morphological analyzer-based methods and neural sequence models.

An alignment-path distillation framework for streaming automatic speech recognition (ASR) with large language models (LLMs) that extracts monotonic alignment paths from a non-streaming teacher's soft text-audio attention and uses them to construct interleaved training sequences.

It is argued that chess provides a useful bridge between cognitive and computational perspectives on strategic reasoning, and identifies explicit planning, grounded and faithful explanation, metacognitive calibration, and human--AI complementarity as directions for future research.

This paper proposes and analyzes Comparison-based Preference Optimization (ComPO), a zeroth-order alignment method based on comparison oracles, and establishes a convergence guarantee for its basic offline scheme under smoothness, gradient sparsity, and compatibility between the oracle and a latent objective.

This survey systematizes the EML landscape by introducing the first structured, model-to-system taxonomy and establishes a structured framework for multimodal systems that are not only high-performing and generalizable but natively efficient and ready for ubiquitous deployment.

Value Flattening is identified as an important yet overlooked failure mode of critic learning in standard PPO and a simple sparse supervision strategy can mitigate it; SParse Proximal Policy Optimization is introduced, which applies the value loss to only a few well-separated states in each response to mitigate both effects.

This work introduces agentic-eCAL, generalizing the Energy Cost of AI Lifecycle (eCAL) metric to directed multi-agent workflows by coupling a closed-form two-rate single-call energy model (compute-bound prefill and memory-bound decode) with 7-layer OSI data transport.

Experimental results show that short browsing sessions are highly identifiable, while future navigation actions are highly predictable from long-term interaction structure combined with recent behavioral context, indicating that repeated interaction patterns remain the dominant predictive signal in the evaluated web-browsing setup.

The findings show that even the presence of a small percentage of biased agents in such a system leads to significant shifts in the opinions of non-biased agents, and semantic analysis demonstrates that rhetorical consistency in textual explanations increases systematically with biased exposure and is partially decoupled from numerical convergenumericalutral agents.

It is shown that epistemic uncertainty leaves a measurable trace within intermediate layers, where spikes in attention entropy are associated with reasoning breakdowns, and that attention dispersion is a promising complement to traditional hallucination detection methods.

Instead of rebuilding the entire recommendation stack from scratch, LIGE-GR generalizes the existing pointwise recommendation system into a listwise generation system, which allows mature recommender systems to benefit from listwise optimization while preserving compatibility with existing models, value functions, and serving infrastructure.

BENCHCOMPASS is introduced, a payment-domain benchmark whose construction pipeline builds scenario-grounded tasks from typed evidence packs, applies LLM-based quality checks, creates task-input attack variants, and reserves final item admission for domain experts.

G-Retriever is competitive with RAG and reduces the error of the base LLM by 72\% with a standard KG and 78\% with a benchmark-aware variant, the gap to RAG narrowing further as the graph is oriented toward task-relevant content.

It is argued that local LLM systems need explicit guarantees for prompt lifetime, persistent storage, and tenant isolation, because the surrounding serving software handles prompt data before, during, and after inference.

Conditioning on outcome class shows that response reliability and correctness are inversely related, whereas genuinely computing methods answered least often and were correct 19.3%.

GeoMesh is presented, a synchronous geo-distributed training framework for heterogeneous GPUs that balances per-worker workloads by assigning each GPU a suitable batch size and number of inner steps, so faster GPUs do more useful work instead of waiting.

HBF memory system HBFlex, a full-HBF memory system with coordinated optimizations for KV reads, writes, and reclamation, is presented, benefiting from higher HBF bandwidth and more efficient management of dynamic KV-cache reads, writes, and erases.

The findings show that even a direction orthogonal to a concept at one layer can contribute to the concept's downstream amplification, and demonstrate that more fine-grained category residuals should also be considered beyond shared general harmfulness representation to fully understand LLM safety.

The Environmental Quality Monitor (EnQyMo) platform is a generic Internet of Things (IoT) middleware designed to process several sensor data related to air quality in indoor spaces and correlate this data with health exposure risks of users/workplace employees.

GYROval - Gridded Yielding of Robust value Orientation is presented, together with the results of administering it to twenty models, a robust benchmark for measuring cultural value orientation in large language models on the two Inglehart-Welzel axes over several domains and roles.

Experiments demonstrate that Code-as-Auditor delivers more accurate and evidence-backed evaluations, enabling automated compliance regulation checking grounded in explicit regulatory criteria and improves the logical consistency of the generated code and the traceability of outcomes.

It is observed that the natural language of problem statements affects LLM-based code generation performance and translation improved Accuracy, but its effectiveness was not consistent across datasets and model types.

This work introduces a Binary Linear Programming model to compute optimal placements and proposes a topology-aware greedy heuristic that efficiently approximates the optimal solution, making it suitable for periodic system reconfigurations.

It is demonstrated how, given a reference implementation, Large Language Models can be utilised for the automated and scalable generation of code that compiles, passes tests, and crucially differs semantically/binary-wise, that can re- place code in the reference implementation, thereby significantly reducing diversification costs.

Encoder Awakening via Adapters (EAVA), a simple yet effective domain-adaptive fine-tuning method for Speech-LLM-based ASR, consistently outperforms vanilla fine-tuning and other baselines, achieving new state-of-the-art performance.

A fine-tuned Small Language Model is fine-tuned with a hybrid ordinal regression loss combining a Mean Squared Error term, which preserves ordinal distance, with a categorical Cross-Entropy term, which sharpens class boundaries, which sharpens class boundaries.

This work proposes SpecLens, an automated framework for LLM-based Verilog generation that derives specification-driven constraints by analyzing behavioral divergence among multiple candidate implementations, using the original specification as the only external semantic source during generation.

Experiments on SemEval-2017 and WebPun show that PunGraph consistently improves the performance of small-scale LLMs and achieves competitive results against strong proprietary models, highlighting the benefits of integrating structured knowledge with LLMs.

The results suggest that model-level alignment is not compositional: individually capable and apparently safe agents can form systems with qualitatively different failure modes as AI becomes persistent and interconnected.

This study proposes the TAISE framework, which repurposes AI weather forecasting models to produce coherent extreme weather sequences at a fraction of traditional costs, and suggests a pathway toward democratising catastrophe risk quantification and enabling dynamic, comprehensive portfolio assessment for insurers, reinsurers, ILS fund managers and public-sector risk managers.

This work proposes an early-bird decoding framework, motivated by the observation that tokens with similarly low entropy tend to cluster and can be jointly decoded earlier, before reaching the confidence threshold, and integrates two key enablers: a learnable network that adaptively groups tokens with similar uncertainty into variable-length blocks, rather than relying on fixed block sizes.