Papers

Augmenting Transformers with linguistic structures effectively enhances the syntactic generalization performance of language models. Previous work in this direction focuses on syntactic tree structures of languages, in particular constituency tree structures. We propose Graph-Infused Layers Transformer Language Model (GiLT) which leverages dependency graphs for augmenting Transformer language models. Unlike most previous work, GiLT does not insert extra structural tokens in language modeling; instead, it injects structural information into language modeling by modulating attention weights in the Transformer with features extracted from the dependency graph that is incrementally constructed along with token prediction. In our experiments, GiLT with semantic dependency graphs achieves better syntactic generalization while maintaining competitive perplexity in comparison with Transformer language model baselines. In addition, GiLT can be finetuned from a pretrained language model to achieve improved downstream task performance. Our code is released at https://github.com/cookie-pie-oops/GiLT-LM.

Continuous diffusion and flow models are attractive for non-autoregressive text generation because they can update all positions in parallel. A major difficulty is the interface between continuous latent states and discrete tokens. This report studies a draft-conditioned latent refinement model built from a frozen BERT encoder, a parallel decoder, a denoising DraftPrior, a local FlowNet, and a learned diagonal MetricNet. Early Gaussian-start experiments showed that good latent-space metrics, such as scale matching or cosine similarity, do not guarantee good decoding. Generated latents can be close to real encoder latents but still produce high-entropy, biased, or repetitive token distributions. We therefore frame the task as controlled local refinement rather than full generation from noise. On ROCStories, using the first two sentences as prompt and the last three as target, full 768-dimensional BERT latents recover tokens much better than compressed 256-dimensional latents. With 768-dimensional latents, DraftPrior target-token probability is 0.938 for clean drafts, 0.613 for 3% token dropout, 0.483 for 5% dropout, and 0.272 for 10% dropout. Local flow refinement and fused decoder-aware readout give modest additional gains, while metric learning and OT-style alignment improve geometry but do not close the decoder gap. The main result is a diagnostic one: latent geometry alone is not enough. Continuous latent text generation should be evaluated by decoder recoverability, the quality of the start distribution, and whether refinement preserves decoder-readable structure.

Distillation enables compact Vision-Language Models (VLMs) to obtain strong reasoning capabilities, yet the prompts driving this process are typically chosen via simple heuristics or aggregated from off-the-shelf datasets. We reveal a critical inefficiency in this approach: up to 69% of the prompts in standard chart / document reasoning datasets are effectively zero-delta, meaning the teacher and student already induce the exact same answer distribution. Training on these prompts provides minimal learning signal, causing student improvement to rapidly saturate regardless of data scale. To escape the zero-delta trap, we return to first principles: distillation fundamentally minimizes distributional divergence, and thus a prompt is valuable only if it exposes a functional capability gap between the teacher and student. We quantify this gap through answer divergence ($\Delta$), demonstrating that non-zero divergence is critical for effective scaling. Building on this insight, we propose a staged synthesis pipeline that repurposes existing datasets as seeds, actively targeting student failure modes to produce better prompts. The result is DeltaPrompts, a diverse dataset of 200k synthetic, high-divergence reasoning problems. We evaluate DeltaPrompts across three distinct settings: on-policy distillation with the target teacher-student pair, transfer to a novel model family without regenerating the data, and off-policy fine-tuning of a non-reasoning model. Across all scenarios, DeltaPrompts drives substantial gains, yielding up to 15% relative improvement even on top of a highly-optimized reasoning model (e.g., Qwen3-VL-8B-Thinking) – averaged over 10 benchmarks spanning chart, document and perception-centric reasoning.

Process Reward Models (PRMs) provide step-level feedback for reasoning, but current PRMs usually output only a single reward score for each step. Downstream methods must therefore treat imperfect step-level reward predictions as reliable decision signals, with no indication of when these predictions should be trusted. We propose BetaPRM, a distributional PRM that predicts both a step-level success probability and the reliability of that prediction. Given step-success supervision from Monte Carlo continuations, BetaPRM learns a Beta belief that explains the observed number of successful continuations through a Beta-Binomial likelihood, rather than regressing to the finite-sample success ratio as a point target. This learned reliability signal indicates when a step reward should be trusted, enabling downstream applications to distinguish reliable rewards from uncertain ones. As one application, we introduce Adaptive Computation Allocation (ACA) for PRM-guided Best-of-N reasoning. ACA uses the learned reliability signal to stop when a high-reward solution is reliable and to spend additional computation on uncertain candidate prefixes. Experiments across four backbones and four reasoning benchmarks show that BetaPRM improves PRM-guided Best-of-N selection while preserving standard step-level error detection. Built on this signal, ACA improves the accuracy–token tradeoff over fixed-budget Best-of-16, reducing token usage by up to 33.57% while improving final-answer accuracy.

The effective detection and governance of Large Language Model (LLM) generated content has become increasingly critical due to the growing risk of misuse. Despite the impressive performance of existing detectors, their reliability and potential in multilingual, real-world scenarios remain largely underexplored. In this study, we introduce DetectRL-X, a comprehensive multilingual benchmark designed to evaluate advanced detectors across 8 dimensions. The benchmark encompasses 8 languages commonly used in commercial contexts and collects human-written texts from 6 domains highly susceptible to LLM misuse. To better aligned with real-world applications, We create LLM-generated texts using 4 popular commercial LLMs, and include typical AI-assisted writing operations such as polishing, expanding, and condensing to capture authentic usage patterns. Furthermore, we develop a multilingual framework for paraphrasing and perturbation attacks to simulate diverse human modifications and writing noise, enabling stress testing of detectors across languages. Experimental results on DetectRL-X reveal the strengths and limitations of current state-of-the-art detectors when applied to diverse linguistic resources. We further analyze how domains, generators, attack strategies, text length, and refinement operations influence performance in different languages, underscoring DetectRL-X as an effective benchmark for strengthening multilingual and language-specific detectors.

We identify intrinsic limitations of Rotary Positional Embeddings (RoPE) in Transformer-based long-context language models. Our theoretical analysis abstracts away from the specific content of the context and depends only on its length. We prove that as context length increases, RoPE-based attention becomes unpredictable and loses two properties that are central to its effectiveness. First, it loses its locality bias: RoPE is no more likely to favor nearer positions than substantially farther ones. Second, it loses consistency in token relevance: a key vector that receives a higher attention score than an alternative at one position may receive a lower score at another. In both cases, the probability of failure approaches 0.5, no better than random guessing. We further prove that the attention score can remain unchanged when a key token is moved to a different position, or even replaced by a different token, indicating a failure to distinguish positions or tokens. Adjusting the RoPE base trades off distinguishing positions against distinguishing tokens but cannot preserve both at the same time. Increasing the RoPE base hyperparameter, a common practice in today's long-context models, helps distinguish different tokens, but inevitably sacrifices the ability to distinguish positions. Our empirical analysis shows that multi-head, multi-layer architectures are insufficient to overcome these limitations. Our findings suggest that fundamentally new mechanisms for encoding position and token order may be needed in future Transformer long-context language models.

The quadratic complexity of attention imposes severe memory and computational bottlenecks on Large Language Model (LLM) inference. This challenge is particularly acute for emerging agentic applications that require processing multi-million token sequences. We propose STS, a sparse attention mechanism that requires no model retraining. STS leverages the key insight that tokens identified as important by a smaller draft model are highly predictive of important tokens for a larger target model. By integrating into speculative decoding frameworks, STS repurposes the draft model's attention scores to dynamically construct a token-and-head-wise sparsity mask. This mask effectively prunes the expensive attention computation in the target LLM. Our evaluation shows that STS achieves a 2.67x speedup operating at approximately 90% sparsity on representative benchmark NarrativeQA, maintaining negligible accuracy degradation compared to dense attention. STS establishes a new state-of-the-art on the sparsity-accuracy trade-off, outperforming prior techniques by enabling higher sparsity levels for a given accuracy budget.

Half a million cuneiform clay tablets survive in museums worldwide, yet modern users can neither read nor write in the world's oldest writing system, leaving a 4,000-year cultural barrier that existing NLP tools have only partially addressed. Prior work enables one-way, scholar-oriented translation from Akkadian to English, but offers no path in the reverse direction: non-specialist users cannot compose new content in cuneiform, and therefore remain passive consumers of ancient culture rather than active participants. We present TabletCraft, the first open-source system that enables bidirectional interaction with Mesopotamian writing. Users can read ancient tablets (Akkadian to English) and compose new messages as cuneiform clay tablets (English to Akkadian to cuneiform to rendered tablet). The system integrates a ByT5-based translation model trained on 116K bidirectional samples, a cuneiform sign converter with 14,240 mappings (95.3% coverage), and a visual tablet renderer, packaged as a pip-installable toolkit with CLI and web demo. On the held-out Akkademia validation split (2,812 samples), we report 49.1 BLEU for Akkadian-to-English and 48.5 BLEU for English-to-Akkadian, the first published quantitative result in the reverse direction.

Inference with large language models (LLMs) on long sequences is computationally expensive due to the quadratic complexity of self-attention. Distributed blockwise methods such as Star Attention reduce this cost by sharding context across hosts, but rely on prepending a static, content-blind copy of the first block to every host. We propose Pulsar Attention, which replaces the static anchor with two lightweight, content-aware components: a small attention-sink prefix that stabilizes softmax, and compact cross-block summaries built via a Max-IDF heuristic that selects chunks containing globally rare tokens. This reduces the Phase 1 per-GPU FLOPs by up to 3.3x over Star Attention while retaining an identical KV cache footprint. On RULER with Llama-3.1-8B-Instruct, Pulsar Attention outperforms Star Attention at sequence lengths up to 128K tokens and remains competitive with dense attention across most tasks, with task-dependent absolute gains of up to 4.7% over the dense baseline.

Large language models excel at code generation for mainstream programming languages but struggle with rare, domain-specific languages such as MiniZinc, a constraint modeling language for combinatorial problems. We investigate whether targeted fine-tuning can teach small language models (0.6B to 20B parameters) to generate syntactically correct and semantically valid MiniZinc models from natural language problem descriptions. Our key finding is that syntax errors dominate failures when working with this domain specific language: the out-of-the-box execution accuracy of small language models such as Qwen3, LLaMa, Gemma, and GPT-OSS is near-zero. We propose a cross-model error bootstrapping approach that collects syntax errors from multiple LLM runs and leverage those to curate an error correction training dataset. This dataset allows us fine-tune small language models that consistently improves both direct code generation and chain-of-thought approaches across all model sizes. With self-reflection and ensembling, our approach achieves up to 98% execution accuracy. In parallel, solution accuracy still remains at 35%, indicating that while syntax is learnable, constraint reasoning remains a challenge. We contribute our fine-tuning pipeline, datasets, and models to opens-source for further research on text-to-model translation.

Human-annotated data remains fundamental to training frontier Large Language Models (LLMs). However, crowd-sourced annotations often suffer from quality issues stemming from annotator misunderstanding or lack of engagement. To address this, we introduce a real-time requirement adherence (RE-AD) framework that leverages LLMs to proactively validate labeling quality. Our methodology involves decomposing Standard Operating Procedures (SOPs) into atomic rules via self-reflection, categorizing them by complexity, and applying tiered validation strategies. Evaluated on a synthetic benchmark, the system achieved an F1 score of 0.749. Furthermore, production deployment resulted in annotators accepting and fixing 82% of the errors flagged by the framework. We include ablation studies to demonstrate the impact of our core design decisions.

All frontier large language models (LLMs) exhibit response drift – producing outputs that deviate from expert-validated references – yet the magnitude and structure of this drift remain uncharacterised by systematic human evaluation. Here we report a fully crossed evaluation in which 47 geographically diverse participants each assessed all 62 multidomain questions across ten frontier LLMs under blinded conditions, yielding 29,140 independent assessments. Every model drifts, but drift magnitude varies substantially: eight models converge on a statistically indistinguishable ceiling (78-81% deviation), while two achieve lower deviation (47-49%). Drift profiles differ across six domains and 62 questions, with pairwise correlations among ceiling models exceeding r = 0.85. Automated similarity metrics explain less than 2% of variance in human judgements. These findings reveal that response drift is universal across frontier LLMs, domain- and question-dependent in structure, and accessible only through human-centred evaluation.

Large language models show promise for clinical prediction, but zero-shot performance on specialized tasks is limited by incomplete domain knowledge, especially for smaller locally deployable models. We present a lightweight knowledge-injection framework for zero-shot ICU delirium prediction that augments a deterministic natural-language summary of structured electronic health record data with an external clinical knowledge report at inference time, without fine-tuning or retrieval. We evaluate LLaMA 3.1 8B and LLaMA 3.3 70B on 3,160 ICU admissions from the MIMIC IV dataset. Adding a clinically meaningful external knowledge report improves AUROC by 8.57 percentage points for the 8B model and 1.99 percentage points for the 70B model compared to no external knowledge. Relative to a GPT-5.2 frontier-model reference without external knowledge report (AUROC 68.86%), knowledge injection reduces the performance gap from 15.66 to 7.09 AUROC points for LLaMA 8B and from 5.30 to 3.31 AUROC points for LLaMA 70B. Random control reports do not improve performance and often degrade it, indicating that gains depend on clinically meaningful content rather than added prompt length alone. SHAP-based attribution further confirms that the injected knowledge is actively used during prediction. These findings suggest that inference-time knowledge injection can narrow the gap between locally deployable open-weight models and frontier closed models while preserving a practical, privacy-preserving workflow for resource-constrained clinical settings.

Semantic Field Theory (SFT) has developed from a philosophical critique of strong anti-formalist readings of language games into a proposed computational model class for lexical semantics, higher order composition, and stabilized interpretation. This paper reconstructs that evolution and gives SFT a sharper mathematical core suitable for independent evaluation in computational linguistics and representation learning. The central proposal is that a tractable level of linguistic organization can be modeled through lexical representations expressed as semantic fields, through contextual deformation of those fields, through interaction terms defined over subsets of tokens, and through stabilization governed by semantic energy dynamics. The paper contributes five formal elements. First, it defines a semantic field model as a tuple consisting of a semantic space, a lexical field lifting, a contextual deformation map, an interaction complex, and an interpretation functional. Second, it proves a Gaussian product closure result showing that multiplicative field interactions have explicit centers, precisions, and compatibility factors. Third, it generalizes the three-word problem by using Mobius inversion on the subset lattice to isolate irreducible semantic interactions of arbitrary order. Fourth, it introduces an order spectrum that measures how much field mass is explained at each interaction order. Fifth, it formulates stabilized interpretation as minimization of an energy functional associated with the sentence and gives existence, descent, and stability conditions. A small worked example shows how a three-word summer day triple can be represented by Gaussian semantic fields, implemented in Python, and summarized by a flow diagram. The result is not a completed theory of natural language meaning and does not replace social, pragmatic, or normative accounts of language.

This paper presents two systems for the GameTox Shared Task at the Workshop on EEUCA at ACL 2026, which requires classifying World of Tanks chat utterances into six fine-grained toxic intent categories (Labels 0-5). Severe class imbalance, domain-specific multilingual slang, and extremely scarce data for rare categories such as Threats (Label 4, 60 samples) and Extremism (Label 5, 24 samples) make this a challenging classification problem. Our primary submission, RAKSHAK (rak s. aka, Sanskrit for "Protector"), is a multi-task DeBERTa-v3-base (He et al., 2022) framework combining rationale distillation from Qwen2.5-14B (An et al., 2024), Supervised Contrastive Loss, and dedicated rare-class binary heads. RAKSHAK's training data is augmented with cross-domain transfer from the Jigsaw Toxic Comment dataset (16,225 samples mapped to Labels 1-4) and 100 LLM-generated extremism samples for Label 5. Our secondary system (M1) fine-tunes DeBERTa-v3-base with Focal Loss on the original GameTox data plus the same 100 extremism samples, without Jigsaw transfer. RAKSHAK achieves a Macro F1 of 0.5883 on the official test set, ranking 7th out of 35 participating teams, while M1 achieves 0.5252 Macro F1. An ablation comparing M1 with and without Jigsaw data shows that cross-domain transfer accounts for +2.6 F1 points, while RAKSHAK's multi-task architecture contributes a further +3.7 points.

When multi-agent LLM systems produce bad answers, not all failures are equal: some answers are grounded in the right material but incomplete, while others are simply ungrounded and should be stopped. Current retry strategies treat both cases identically (try again and hope for the best), leaving human supervisors unable to tell whether a retry was warranted or whether the system should have halted instead. We introduce the Argent Signaling Protocol (ASP), a compact machine-readable header that accompanies every AI-generated response with structured quality signals: certainty (@C), grounding (@G), stochasticity (@S), and an assumption index that classifies the evidentiary basis of each claim. These signals enable a controller to distinguish repairable failures from containment failures and route each case differently. We evaluate ASP in two modes. In standalone mode, a 27-question document-grounded QA benchmark over the Array BioPharma/Ono license agreement compares baseline prompts against ASP-instrumented controller actions across three local GGUF models. On Qwen~(0.8B), ASP improves pass rate from 11.1% to 33.3% and mean term coverage from 36.7% to 65.4%; on Dobby~(8B), ASP produces 4 fail-to-pass recoveries, raising pass rate from 33.3% to 44.4%; on SmolLM3~(3B), ASP alternates between repair and containment per question. Aggregate improvement is meaningful (12/81 to 21/81 passes). In multi-agent mode, an ASP sidecar sits between a retrieval agent and a downstream decision agent; the sidecar blocks 100% of ungrounded upstream outputs from reaching the downstream agent (24/27 blocked, 0 ungrounded propagations).

Forecasting human life outcomes is important to gain insights into how individuals attain long and healthy lives. Conventional statistical approaches yield limited accuracy, potentially due to discarding the sequential structure of the life course. Modern methods such as transformer architectures require large scale training data that most longitudinal panel studies lack. Here we introduce LifeSentence, a model for life-course reasoning that bridges large language models with longitudinal panel data. By representing each life event as a structured natural-language record and instruction-tuning a pretrained 24-billion-parameter language model across an 18-task evaluation taxonomy spanning prediction, robustness and reasoning, LifeSentence supplements panel data with distributional knowledge already encoded during pretraining. Trained on approximately 65,000 individuals from the German Socio-Economic Panel - roughly 45 times fewer than prior transformer-based approaches - LifeSentence outperforms classical and deep learning baselines across all task families, achieving a threefold improvement in joint event-and-timing prediction from best baselines and 91.2% Kendall's tau when reconstructing chronological order from timestamp-stripped event sets. Without explicit supervision, the model recovers documented patterns of social stratification, including the education premium, the gender wage gap and the motherhood penalty, from discrete event sequences alone. A natural-language interface further enables qualitatively new research queries, such as connecting an early-life history to a specified late-life endpoint, establishing LifeSentence as both a predictive tool and a probe for counterfactual exploration of human biographies.

Large language models (LLMs) are increasingly being applied to financial analysis, reporting, investment decision support, risk management, compliance, and professional training. However, robust evaluation of their domain competence in finance remains incomplete. Widely used open benchmarks such as FinQA, ConvFinQA, and TAT-QA have played an important role in advancing financial question answering and numerical reasoning, but they focus primarily on question answering over financial reports and do not provide an explicit hierarchy of professional difficulty. Broader resources, including FinanceBench, PIXIU, FinBen, and FLaME, expand the coverage of financial tasks, yet the problem of evaluating the transition from foundational knowledge to expert-level financial reasoning remains open. In this work, we present FINESSE-Bench, a suite of eight specialized benchmarks comprising 3,993 questions for hierarchical evaluation of financial competencies in LLMs. FINESSE-Bench combines exam-oriented datasets inspired by professional certifications (CFA-like Levels 1-3, CMT-like Level 2, and CFTe-like Level 1), applied trading task collections, and a Russian-language olympiad benchmark. This design enables evaluation of domain breadth, performance degradation as difficulty increases, the ability to solve computational tasks, and model behavior in specialized financial domains. We also describe a unified evaluation protocol covering multiple-choice questions, numerical answers, and short open-ended responses, together with an automated scoring scheme for freeform answers based on the LLM-as-judge paradigm. FINESSE-Bench is intended both as a complement to existing open financial benchmarks and as a tool for more substantive evaluation of professionally relevant financial competencies in large language models.

Conversational nurse-patient transcripts contain actionable observations, but converting these transcripts into structured representations at scale remains challenging. Documentation burden is substantial, with prior studies showing clinicians spend large portions of their workday on documentation and related desk work rather than direct patient care. MEDIQA-SYNUR focuses on observation extraction from conversational nurse-patient transcripts, requiring systems to normalize these narratives into a predefined schema with value-type constraints. We propose a modular retrieval-augmented generation (RAG) pipeline that uses the training set as an exemplar corpus, combines schema-constrained prompting (full schema vs. pruned candidate schema), deterministic schema-based postprocessing, and a second-pass audit, with two LLM backbones: Llama-4-Scout-17B-16E-Instruct and GPT-5.2 with corresponding embedding models for RAG. Our best configuration uses GPT-5.2 with full schema, RAG, and a second-pass auditing, achieving 80.36% F1 score. Overall, our results show that RAG consistently improves performance, while the optimal degree of schema constraint depends on the model, and second-pass auditing yields modest additional gains by correcting residual schema-adherence errors.

Reinforcement learning with verifiable rewards (RLVR) is widely used to improve the reasoning capability of language models, since model-generated answers can be easily verified by a rule-based verifier. In contrast, reinforcement learning from human feedback (RLHF) is mainly used to optimize open-ended conversation with a learned reward model, because such responses are not easy to verify. RLVR and RLHF are usually trained separately with their own domain data and training protocols. How gains transfer between these domains is unclear, partly because comparisons change both where a model is trained (the prompts) and what is rewarded (by a verifier or a reward model). We separate the two in a controlled study of RL on multiple base models that fixes the optimizer and the number of training prompts, and varies only the source domain of the prompts (math, multi-domain, chat, or mixtures) and the reward. We find that RLHF on open-ended prompts substantially improves conversational ability and also brings large gains in reasoning and coding, whereas RLVR improves reasoning but its gains in open-ended conversation plateau far below those of RLHF, suggesting that RL on open-ended prompts with a reward model can serve as an effective approach for generalizing across domains.

Reasoning-trained language models often spend more tokens on harder problems, but longer chains of thought do not show whether a model is merely computing for more steps or following a different internal trajectory. We study this distinction through hidden-state trajectories during chain-of-thought generation across competitive programming, mathematics, and Boolean satisfiability. Raw trajectory geometry is strongly shaped by generation length: longer generations mechanically alter path statistics, so difficulty-dependent comparisons are misleading without adjustment. After residualizing trajectory statistics on length, difficulty remains systematically coupled to corrected trajectory geometry across all domains studied. The clearest reasoning-specific separation appears in the code domain, where harder problems show more direct corrected trajectories and less heterogeneous local curvature in reasoning-trained models than in matched instruction-tuned baselines. Corrected difficulty-geometry coupling is weaker, but still present, in mathematics and Boolean satisfiability. Prompt-stage linear probes do not mirror the code-domain separation, and behavioral annotations show that stronger corrected coupling co-occurs with strategy shifts and uncertainty monitoring. Together, these findings establish length correction as a prerequisite for generation-time trajectory analysis and show that reasoning training can be associated with distinct corrected trajectory geometry, with the strength of the effect depending on the domain.

Surprisal theory posits that the processing difficulty of a word is determined by its predictability in context, offering a potential link between human sentence processing and next-word predictions from language models. While language model (LM) surprisals successfully predict reading times in naturalistic text, they systematically underpredict the magnitude of difficulty observed in controlled studies of syntactic ambiguity, particularly in garden path sentences. This mismatch might arise from differences in the computational constraints between humans and LMs. Here we test one such hypothesis, specifically, that LMs may be able to simultaneously consider a greater number of distinct sentence interpretations at once, compared to humans. Using Recurrent Neural Network Grammars (RNNGs) with word-synchronous beam search, we systematically vary the number of simultaneous parses used to compute word surprisal, and then use these surprisals to predict human reading times. Reducing the number of simultaneous active parses indeed increases the magnitude of predicted garden path effects, but not nearly enough to capture the full magnitude of the effects in humans. This suggests that differences in the number of simultaneous parses available to LMs and humans cannot reconcile LM-based surprisal with human sentence processing.

This paper presents a comprehensive analysis of neural activation patterns across six distinct large language model (LLM) architectures, examining their performance on twelve cognitive task categories. Through systematic measurement of final activation values, attention entropy, and sparsity patterns, we reveal fundamental differences in how encoder and decoder architectures process diverse cognitive tasks. Our analysis of 144 task-model combinations demonstrates that mathematical reasoning consistently produces the highest attention entropy across all architectures, while decoder models exhibit significantly higher sparsity patterns compared to encoder models. The findings provide critical insights into the computational characteristics of modern language models and their task-specific neural behaviors, with implications for model selection and optimization in big data applications.

Modern quantitative trading increasingly relies on systematic models to extract predictive signals from large-scale financial data, where alpha factor discovery plays a central role in transforming market observations into tradable signals. Recent LLM-based methods have shown promise in automating factor generation, but most of them still rely on prompt-level generation–evaluation–feedback loops for iterative optimization. As the loop becomes longer, repeatedly appended historical candidates and feedback can cause context explosion, increase inference cost, dilute useful information, and introduce feedback drift. Moreover, these methods often depend on very large LLMs whose stable generation preferences may lead to structurally similar expressions, redundant candidates, and search stagnation. To address these limitations, we propose QuantEvolver, a self-evolving alpha factor discovery framework based on reinforcement fine-tuning. Instead of accumulating feedback in the prompt, QuantEvolver converts executable quantitative evaluation into policy updates, enabling a Miner LLM to internalize historical optimization experience through parameter learning. Specifically, QuantEvolver constructs high-quality seed factors, builds diverse seed–time-window training tasks, generates executable Factor DSL expressions, evaluates them through Regime Backtest, and optimizes the Miner LLM with Diversity-Complementarity Reward. During training, high-quality factors are continuously accumulated in a Mined Factor Database, which serves as the final discovered factor library. Extensive experiments on three realistic market benchmarks demonstrate the effectiveness of QuantEvolver, which consistently improves the primary evaluation metric of each task over existing LLM-based alpha factor discovery baselines, produces higher-quality and more complementary factor pools.

Large language model personalization typically adapts outputs to user preferences and style but does not account for differences in user evaluation capacity across domains of expertise. This limitation can encourage Professional Domain Drift, where users rely on AI generated reasoning in domains they cannot reliably evaluate. We introduce Capability Conditioned Scaffolding, a typed framework that partitions expertise into strong, mixed, and weak domains and conditions intervention behavior on structured capability profiles. A pilot evaluation across multiple MMLU subsets and four LLM substrates shows consistent profile conditioned intervention behavior, including categorical inversion under profile swapping and selective activation in mixed domain risk zones. These findings suggest that capability aware scaffolding can support more reliable professional human AI collaboration beyond stylistic personalization.

Recent advances in generative AI have shown their potential to be leveraged for legal education. Yet, work on the development and deployment of such systems for legal education in the Global South is limited. In this work, we developed Eskwai for Students, a generative AI assistant to help law students with their legal education. Eskwai for Students is a retrieval augmented generation (RAG) system that provides answers to a wide range of legal questions for law students grounded in a curated database of over 12K case laws and 1.4K legislation in Ghana. We deployed Eskwai for Students in a longitudinal study of 30 months (2.5 years) used by 3.1K law students in Ghana who made 32K queries. We evaluated the helpfulness of our AI, and provided insight into the kinds of queries law students submit to this generative AI tool, which raises some ethical concerns. This work contributes to an understanding of how law students in the Global South are using generative AI for their studies and the ways it could be leveraged responsibly to advance legal education.

Sub-Saharan Africa faces persistently high student-teacher ratios and shortages of qualified teachers, limiting students' access to personalized learning support and formative assessment. To address this challenge, we present Adesua, a WhatsApp-based AI Teaching Assistant for science education that extends the Kwame for Science platform. Adesua leverages WhatsApp's widespread adoption in Africa to provide accessible, curriculum-aligned learning support for Junior High School (JHS) and Senior High School (SHS) students across West Africa. The system integrates curated textbooks and 33 years of national examination questions with generative AI to enable conversational question answering and automated assessment with feedback via a WhatsApp bot. Students can ask science questions, take timed or untimed multiple-choice tests by topic or exam year, and receive instant grading and detailed explanations of correct and incorrect responses. A 6-month feasibility deployment in 2025 had 56 active users in Ghana, including students and parents. Quantitative evaluation showed a high perceived usefulness, with a helpfulness score of 93.75% for AI-generated answers, albeit with a small number of ratings (n=16). These preliminary results provide a basis for more extensive future evaluation of a WhatsApp-based AI assistant to assess its potential to offer scalable, low-cost personalized learning support and formative assessment in resource-constrained educational contexts.

Communicating using only a limited vocabulary is a common but challenging cognitive phenomenon, requiring an ideal communicator to plan carefully to optimize for intelligibility while circumventing a constrained lexicon. In this work, we investigate how humans respond to a broad array of questions under variable vocabulary limitations, consisting of only 250 highly frequent words at the most restrictive. We provide theoretically motivated comparisons to greedy and globally optimal sampling algorithms using Sequential Monte Carlo inference with large language models. Humans generally resemble greedy sampling more than globally optimal sampling, though more skilled humans are more likely to backtrack and revise – a non-greedy behavior. An observed human pattern of leaning on semantically light words in high-constraint settings falls out of both greedy and globally optimal sampling. We discuss the results and their broader implications for resource-rational cognition, psycholinguistics, L2 communication, and language impairments.

Half a billion citation edges extracted from 100.7 million Ukrainian court decisions reveal that judicial citation structure encodes legal domain boundaries without supervision and predicts future legislative importance with near-perfect accuracy. We construct the first large-scale citation graph from the complete EDRSR registry (99.5 million full texts, 1.1 TB), extracting 502 million citation links across six types via regex on commodity hardware in approximately 5 hours, with precision of 1.00 on a 200-decision validation sample (95% Wilson CI: [0.982, 1.000]). Three principal findings emerge. (1) The degree distribution follows a power law (alpha = 1.57 +/- 0.008), placing the Ukrainian court network near the EU Court of Justice and below the US Supreme Court, with hub articles cited by millions of decisions. (2) Louvain community detection on the co-citation projection recovers legal domain boundaries (civil, criminal, administrative, commercial) with modularity Q = 0.44-0.55 and temporal stability (NMI = 0.83-0.86 across periods), constituting an automatically constructed legal ontology grounded in judicial practice. (3) Citation features predict top-1000 articles with AUC = 0.9984, substantially outperforming a naive frequency baseline (P@1000 = 0.655); temporal dynamics detect legislative regime changes as phase transitions and the 2022 invasion as a citation entropy spike (H: 11.02 -> 13.49) with emergent wartime legislation nodes. The citation-derived ontology is operationalized as the domain layer of a workflow memory system for LLM-assisted legal analysis, connecting to the ontology-controlled paradigm. The extraction pipeline, analysis code, and aggregated statistics are released as open data.

The automatic synthesis of a program from any form of specification is regarded as a holy grail of computer science. Fueled by LLMs, NL2Code has achieved tremendous success, yet the fundamentally more challenging task of synthesizing programs from input-output behavior, which we refer to as IO2Code, remains largely unsolved. Whereas NL2Code can exploit the semantic alignment between natural language and code acquired during pretraining, IO2Code requires recovering underlying principles from concrete computational behavior, navigating a vast and underspecified hypothesis space. To address this, we propose DIO-Agent, a discovery agent that performs agentic reverse engineering from input-output behavior to code. Our method frames IO2Code as an evolutionary search over discrete program space, in which an LLM serves as the mutation operator and concrete error signals from execution guide each mutation. To prevent the search from wandering into structurally complex yet incorrect dead ends, we introduce the Transformation Priority Premise as a mutation prior that biases the LLM toward the simplest hypothesis consistent with current evidence, progressively escalating from constants to conditionals to iteration only when simpler constructs are insufficient. To facilitate systematic study, we further construct an IO2CodeBench spanning multiple difficulty levels. Extensive experiments show that DIO-Agent consistently outperforms both traditional program-by-example method and SOTA evolution-agent baselines across all difficulty levels and various LLMs, while substantially surpassing test-time scaling strategies with equivalent sampling budgets.