Language Modeling

Follow this topic Ask about this topic

Language modeling studies how systems learn from sequences, most often by predicting the next token or reconstructing masked tokens, enabling text generation and language understanding. Common approaches include autoregressive Transformers, masked language models, and alternatives such as recurrent or state-space networks, mixture-of-experts models, and diffusion methods. Researchers improve models through data and tokenizer design, continued pretraining, instruction tuning, retrieval, distillation, and compression, evaluating them with perplexity, linguistic tests, and task benchmarks. A central tension is how to make models more efficient and reliable while ensuring benchmark success reflects robust reasoning, generalization, and competence across languages and domains.

Latest papers

525 papers73 papers in the last four weeks, up 22% on the four weeks before. 0.5% of all new papers.

Mar 3, 2026

Raising Bars, Not Parameters: LilMoo Compact Language Model for Hindi

The dominance of large multilingual foundation models has widened linguistic inequalities in Natural Language Processing (NLP), often leaving low-resource languages underrepresented. This paper introduces LilMoo, a 0.6-billion-parameter Hindi language model trained entirely from scratch to address this gap. Unlike prior Hindi models that rely on continual pretraining from opaque multilingual foundations, LilMoo is developed through a fully transparent and reproducible pipeline optimized for limited compute environments. We construct a high-quality Hindi corpus (GigaLekh) filtered through both heuristic and learned (LLM-as-a-judge) methods, complemented by bilingual augmentation with curated English data. Using this dataset, we explore various training recipes for small-scale language models. Across comprehensive evaluation suites, LilMoo consistently outperforms comparably sized multilingual baselines such as Qwen2.5-0.5B and Qwen3-0.6B, demonstrating that well-designed language-specific pretraining can rival large multilingual models at the sub-billion-parameter range.

Shiza Fatimah, Aniket Sen, Sophia Falk +3

Bonn-Aachen International Center for Information Technology (b-it) / CAISA Lab · Lamarr Institute for Machine Learning and Artificial Intelligence · Helmholtz-Institut für Strahlen- und Kernphysik +1

Language Modeling Language Model Pretraining Small Language Models Low-Resource Language Processing

Mar 2, 2026

Recursive Models for Long-Horizon Reasoning

Modern language models reason within bounded context, an inherent constraint that poses a fundamental barrier to long-horizon reasoning. We identify recursion as a core principle for overcoming this barrier, and propose recursive models as a minimal realization, where the model can recursively invoke itself to solve subtasks in isolated contexts. We prove that any computable problem admits a recursive decomposition of reasoning in which each subtask requires only exponentially smaller active context than standard autoregressive models; this strictly surpasses any context management approach confined to a single sequence, such as summarization. We further generalize our framework to modern agentic systems with arbitrary context processing and control flows, and prove that recursive models can achieve optimal power within this broader class. Experimentally, we test two settings: fine-tuning a pretrained base model for recursive SAT solving, and training a small model from scratch on Go traces generated by exact game-tree search. Both show improved long-horizon accuracy with small active contexts.

Chenxiao Yang, Nathan Srebro, Zhiyuan Li

Toyota Technological Institute at Chicago

Recursive Language Models Language Modeling Algorithmic Reasoning LLM Reasoning Long-Horizon Reasoning Long-Context Modeling+3

Mar 2, 2026

AdaPonderLM: Gated Pondering Language Models with Token-Wise Adaptive Depth

Test-time scaling via recurrent/iterative Transformers enables large language models to spend more computation at inference, but most pretrained recurrent LMs run a fixed number of iterations, wasting compute on easy tokens and lacking token-wise adaptivity. Following the core idea of Adaptive Computation Time(ACT) and Early Exit(EE), we propose AdaPonderLM, a self-supervised recurrent language model that learns token-wise early exiting during pretraining without manually tuned per-token/per-layer pruning ratios. AdaPonderLM uses iteration-specific MLP gates with a monotonic halting mask to decide when each token stops recurring, and introduces a KV reuse mechanism that reuses cached key/value states for halted tokens, ensuring train--test consistency and practical acceleration. Across Pythia backbones from 70M to 410M (pretraining) and up to 2.8B (continued pretraining), AdaPonderLM reduces inference compute at about 10% while maintaining comparable language modeling perplexity and competitive downstream accuracy. Our analysis shows the learned gates allocate more computation to high-NLL (hard) tokens, exhibiting adaptive computation time behavior in a fully self-supervised setting. Meanwhile, under iso-FLOPs, the learned halting policy consistently outperforms fixed pruning, showing AdaPonderLM allocates compute to the right tokens rather than just reducing average depth.

Shixiang Song, He Li, Zitong Wang +6

LUMIA Lab, School of Artificial Intelligence, Shanghai Jiao Tong University · Shanghai Innovation Institute · Sun Yat-sen University +2

Recurrent Language Models Inference-Time Adaptation of Language Models LLM Inference Acceleration Language Modeling Adaptive Computation

Mar 2, 2026

D3LM: A Discrete DNA Diffusion Language Model for Bidirectional DNA Understanding and Generation

Early DNA foundation models adopted BERT-style training, achieving good performance on DNA understanding tasks but lacking generative capabilities. Recent autoregressive models enable DNA generation, but employ left-to-right causal modeling that is suboptimal for DNA where regulatory relationships are inherently bidirectional. We present D3LM (Discrete DNA Diffusion Language Model), which unifies bidirectional representation learning and DNA generation through masked diffusion. D3LM directly adopts the Nucleotide Transformer (NT) v2 architecture but reformulates the training objective as masked diffusion in discrete DNA space, enabling both bidirectional understanding and generation capabilities within a single model. Compared to NT v2 of the same size, D3LM achieves improved performance on understanding tasks. Notably, on regulatory element generation, D3LM achieves an SFID of 10.92, closely approaching real DNA sequences (7.85) and substantially outperforming the previous best result of 29.16 from autoregressive models. Our work suggests diffusion language models as a promising paradigm for unified DNA foundation models. We further present the first systematic study of masked diffusion models in the DNA domain, investigating practical design choices such as tokenization schemes and sampling strategies, thereby providing empirical insights and a solid foundation for future research. D3LM has been released at https://huggingface.co/collections/Hengchang-Liu/d3lm.

Zhao Yang, Hengchang Liu, Chuan Cao +1

Gaoling School of Artificial Intelligence, Renmin University of China · Zhongguancun Academy

Molecular Foundation Models Genomic Foundation Models Language Modeling Molecular Generation Diffusion Language Models Masked Diffusion Models+1

Mar 2, 2026

Constructing Synthetic Instruction Datasets for Improving Reasoning in Domain-Specific LLMs: A Case Study in the Japanese Financial Domain

In adapting LLMs to specific domains, achieving both domain expertise and reasoning ability remains an urgent challenge. This study proposes a general method for constructing high-quality synthetic instruction data for any domain, starting from domain-specific vocabulary. As a demonstration, we applied this method to the financial domain and constructed a large-scale instruction dataset totaling approximately 9.5 billion tokens with Chain-of-Thought reasoning traces. Evaluation results confirmed performance improvements over baseline models on financial benchmarks, demonstrating the effectiveness of our approach. We also report findings on the impact of reasoning trace length on performance and its limitations. Lastly, we open-source our models and datasets on https://huggingface.co/nri-ai .

Yuma Okochi, Fabio Milentiansen Sim, Tomoyasu Okada

Nomura Research Institute, Ltd. · NRI Indonesia

Financial Language Model Evaluation Financial Reasoning in Language Models Synthetic Data Generation for Language Models Domain Adaptation for LLMs Language Modeling

Feb 28, 2026

A Typologically Grounded Evaluation Framework for Word Order and Morphology Sensitivity in Multilingual Masked LMs

We introduce a typology-aware diagnostic for multilingual masked language models that tests reliance on word order versus inflectional form. Using Universal Dependencies, we apply inference-time perturbations: full token scrambling, content-word scrambling with function words fixed, dependency-based head--dependent swaps, and sentence-level lemma substitution (+L), which lemmatizes both the context and the masked target label. We evaluate mBERT and XLM-R on English, Chinese, German, Spanish, and Russian. Full scrambling drives word-level reconstruction accuracy near zero in all languages; partial and head--dependent perturbations cause smaller but still large drops. +L has little effect in Chinese but substantially lowers accuracy in German/Spanish/Russian, and it does not mitigate the impact of scrambling. Top-5 word accuracy shows the same pattern: under full scrambling, the gold word rarely appears among the five highest-ranked reconstructions. We release code, sampling scripts, and balanced evaluation subsets; Turkish results under strict reconstruction are reported in the appendix.

Anna Feldman, Libby Barak, Jing Peng

Montclair State University New Jersey, USA

Multilingual Language Model Evaluation Language Modeling Masked Language Modeling

Feb 27, 2026

How Large Language Models Get Stuck: Early structure with persistent errors

Linguistic insights may help make Large Language Model (LLM) training more efficient. We trained Meta's OPT model on the 100M word BabyLM dataset, and evaluated it on the BLiMP benchmark, which consists of 67 classes, each defined by sentence pairs that differ in a targeted syntactic or semantic rule violation. We tested the model's preference for grammatical over ungrammatical sentences across training iterations and grammatical types. In nearly one-third of the BLiMP classes, OPT fails to consistently assign a higher likelihood to grammatical sentences, even after extensive training. When it fails, it often establishes a clear (erroneous) separation of the likelihoods at an early stage of processing and sustains this to the end of our training phase. We hypothesize that this mis-categorization is costly because it creates entrenched biases that must, eventually, be reversed in order for the model to perform well. We probe this phenomenon using a mixture of qualitative (based on linguistic theory and the theory of Deep Learning) and quantitative (based on numerical testing) assessments. Our qualitative assessments indicate that only some BLiMP tests are meaningful guides. We conclude by articulating a hypothesis, the Bigram Hypothesis, which claims that the learning process will exhibit erroneous entrenchment if bigram statistics bias the model toward wrong distinctions early in training, and we describe a method of testing the hypothesis on appropriately selected BLiMP classes.

Alokesh Manna, William Snyder, Whitney Tabor

Department of Statistics University of Connecticut, Storrs, USA · Department of Linguistics University of Connecticut, Storrs, USA · Department of Psychological Sciences University of Connecticut, Storrs, USA

Language Modeling LLM Training Language Model Generation Evaluation

Feb 27, 2026

Monotropic Artificial Intelligence: Toward a Cognitive Taxonomy of Domain-Specialized Language Models

The prevailing paradigm in artificial intelligence research equates progress with scale: larger models trained on broader datasets are presumed to yield superior capabilities. This assumption, while empirically productive for general-purpose applications, obscures a fundamental epistemological tension between breadth and depth of knowledge. We introduce the concept of Monotropic Artificial Intelligence -- language models that deliberately sacrifice generality to achieve extraordinary precision within narrowly circumscribed domains. Drawing on the cognitive theory of monotropism developed to understand autistic cognition, we argue that intense specialization represents not a limitation but an alternative cognitive architecture with distinct advantages for safety-critical applications. We formalize the defining characteristics of monotropic models, contrast them with conventional polytropic architectures, and demonstrate their viability through Mini-Enedina, a 37.5-million-parameter model that achieves near-perfect performance on Timoshenko beam analysis while remaining deliberately incompetent outside its domain. Our framework challenges the implicit assumption that artificial general intelligence constitutes the sole legitimate aspiration of AI research, proposing instead a cognitive ecology in which specialized and generalist systems coexist complementarily.

Antonio de Sousa Leitão Filho, Allan Kardec Duailibe Barros Filho, Fabrício Saul Lima +2

Aia Context, São Luís, MA, Brazil · Federal University of Maranhão, São Luís, MA, Brazil

Structural Engineering Language Modeling Cognitive Architectures for AI Agents

Feb 27, 2026

The GRADIEND Python Package: An End-to-End System for Gradient-Based Feature Learning

We present gradiend, an open-source Python package that operationalizes the GRADIEND method for learning feature directions from factual-counterfactual MLM and CLM gradients in language models. The package provides a unified workflow for feature-related data creation, training, evaluation, visualization, persistent model rewriting via controlled weight updates, and multi-feature comparison. We demonstrate gradiend through an English pronoun running example, a semantic sentiment use case that evaluates lexical generalization to held-out target words, and a large-scale feature comparison.

Jonathan Drechsel, Steffen Herbold

Faculty of Computer Science and Mathematics, University of Passau

LLM Interpretability Language Modeling Language Model Editing

Feb 27, 2026

From Tokens To Agents: A Researcher's Guide To Understanding Large Language Models

Researchers face a critical choice: how to use -- or not use -- large language models in their work. Using them well requires understanding the mechanisms that shape what LLMs can and cannot do. This chapter makes LLMs comprehensible without requiring technical expertise, breaking down six essential components: pre-training data, tokenization and embeddings, transformer architecture, probabilistic generation, alignment, and agentic capabilities. Each component is analyzed through both technical foundations and research implications, identifying specific affordances and limitations. Rather than prescriptive guidance, the chapter develops a framework for reasoning critically about whether and how LLMs fit specific research needs, finally illustrated through an extended case study on simulating social media dynamics with LLM-based agents.

Daniele Barolo

Language Modeling LLM Alignment Multi-Agent LLMs

Feb 27, 2026

SleepLM: Natural-Language Intelligence for Human Sleep

We present SleepLM, a family of sleep-language foundation models that enable human sleep alignment, interpretation, and interaction with natural language. Despite the critical role of sleep, learning-based sleep analysis systems operate in closed label spaces (e.g., predefined stages or events) and fail to describe, query, or generalize to novel sleep phenomena. SleepLM bridges natural language and multimodal polysomnography, enabling language-grounded representations of sleep physiology. To support this alignment, we introduce a multilevel sleep caption generation pipeline that enables the curation of the first large-scale sleep-text dataset, comprising over 100K hours of data from more than 10,000 individuals. Furthermore, we present a unified pretraining objective that combines contrastive alignment, caption generation, and signal reconstruction to better capture physiological fidelity and cross-modal interactions. Extensive experiments on real-world sleep understanding tasks verify that SleepLM outperforms state-of-the-art in zero-shot and few-shot learning, cross-modal retrieval, and sleep captioning. Importantly, SleepLM also exhibits intriguing capabilities including language-guided event localization, targeted insight generation, and zero-shot generalization to unseen tasks. All code and data will be open-sourced.

Zongzhe Xu, Zitao Shuai, Eideen Mozaffari +3

University of California, Los Angeles

Multimodal Pretraining Cross-Modal Retrieval Language Modeling

Feb 25, 2026

A Diversity Diet for a Healthier Model: A Case Study of French ModernBERT

Diversity has been gaining interest in the NLP community in recent years. At the same time, state-of-the-art transformer models such as ModernBERT use very large pre-training datasets, which are driven by size rather than by diversity. This summons to investigate the impact of diversity on pre-training. We do so in this study, with the express intent of reducing pre-training dataset size, while retaining at least comparable performance. We compare diversity-driven sampling algorithms, and we use the best one to pre-train several ModernBERT models on French with a fixed compute budget. We fine-tune and evaluate them on a variety of French benchmarks. We compare them with models pre-trained on randomly sampled data of commensurate size, with the same compute budget. We find that both random and diversity-driven sampling may reduce the pre-training dataset by up to 94% and the pre-training time by up to 73% while maintaining performance. Moreover, in some tasks, the inherent quality of models, estimated via head-only fine-tuning, is up to 10 points higher with diversity sampling than with random sampling.

Louis Estève, Christophe Servan, Thomas Lavergne +1

Université Paris-Saclay, CNRS, LISN

Language Model Pretraining Diversity-Aware Sampling Efficient Language Model Training Language Modeling

Feb 25, 2026

Enhancing Legal LLMs through Metadata-Enriched RAG Pipelines and Direct Preference Optimization

Large Language Models (LLMs) perform well in short contexts but degrade on long legal documents, often producing hallucinations such as incorrect clauses or precedents. In the legal domain, where precision is critical, such errors undermine reliability and trust. Retrieval Augmented Generation (RAG) helps ground outputs but remains limited in legal settings, especially with small, locally deployed models required for data privacy. We identify two failure modes: retrieval errors due to lexical redundancy in legal corpora, and decoding errors where models generate answers despite insufficient context. To address this, we propose Metadata Enriched Hybrid RAG to improve document level retrieval, and apply Direct Preference Optimization (DPO) to enforce safe refusal when context is inadequate. Together, these methods improve grounding, reliability, and safety in legal language models.

Suyash Maniyar, Deepali Singh, Rohith Reddy

LLM Grounding Language Modeling RAG Robustness Legal QA LLM Safety

Feb 24, 2026

Deep Sequence Modeling with Quantum Dynamics: Language as a Wave Function

We introduce a sequence modeling framework in which the latent state is a complex-valued wave function evolving on a finite-dimensional Hilbert space under a learned, time-dependent Hamiltonian. Unlike standard recurrent architectures that rely on gating mechanisms to suppress competing hypotheses, our framework utilizes quantum interference: the Hamiltonian steers the phases of complex amplitudes so that conflicting interpretations cancel while compatible ones reinforce. The dynamics are strictly unitary, ensuring that the state norm is preserved exactly at every time step via a Cayley (Crank--Nicolson) discretization. Token probabilities are extracted using the Born rule, a quadratic measurement operator that couples magnitudes and relative phases. Our primary theoretical contribution is a separation theorem characterizing the representational advantage of this readout: we define a family of disambiguation tasks that a complex unitary model of dimension

N

solves exactly, but which requires a state dimension of

Ω(N^2)

for any real-valued orthogonal model equipped with a standard affine-softmax readout. This quadratic gap arises because the Born rule implicitly lifts the

N

-dimensional state into the space of rank-one Hermitian matrices, accessing pairwise phase correlations that are inaccessible to linear projections. Finally, we derive a continuity equation for the latent probability mass, yielding conserved pairwise currents that serve as a built-in diagnostic for tracing information flow between dimensions.

Ahmed Nebli, Hadi Saadatdoorabi, Kevin Yam

cAI Technology GmbH

Sequence Modeling Quantum-Inspired ML Language Modeling Quantum Machine Learning Quantum Neural Networks

Feb 20, 2026

Subgroups of $U(d)$ Induce Natural RNN and Transformer Architectures

This paper presents a direct framework for sequence models with hidden states on closed subgroups of U(d). We use a minimal axiomatic setup and derive recurrent and transformer templates from a shared skeleton in which subgroup choice acts as a drop-in replacement for state space, tangent projection, and update map. We then specialize to O(d) and evaluate orthogonal-state RNN and transformer models on Tiny Shakespeare and Penn Treebank under parameter-matched settings. We also report a general linear-mixing extension in tangent space, which applies across subgroup choices and improves finite-budget performance in the current O(d) experiments.

Joshua Nunley

Cognitive Science Program & Department of Informatics, Indiana University

Recurrent Neural Networks Transformer Language Modeling

Feb 19, 2026

On the scaling relationship between cloze probabilities and language model next-token prediction

Recent work has shown that larger language models have better predictive power for eye movement and reading time data. While even the best models under-allocate probability mass to human responses, larger models assign higher-quality estimates of next tokens and their likelihood of production in cloze data because they are less sensitive to lexical co-occurrence statistics while being better aligned semantically to human cloze responses. The results provide support for the claim that the greater memorization capacity of larger models helps them guess more semantically appropriate words, but makes them less sensitive to low-level information that is relevant for word recognition.

Cassandra L. Jacobs, Morgan Grobol

Department of Linguistics, University at Buffalo, Buffalo, NY, USA · MoDyCo, Université Paris Nanterre, Paris, France

Language Modeling Language Model Scaling Computational Psycholinguistics LLM Evaluation

Feb 19, 2026

ArXiv-to-Model: A Practical Study of Scientific LM Training

While frontier large language models demonstrate strong reasoning and mathematical capabilities, the practical process of training domain-specialized scientific language models from raw sources remains under-documented. In this work, we present a detailed case study of training a 1.36B-parameter scientific language model directly from raw arXiv LaTeX sources spanning mathematics, computer science, and theoretical physics. We describe an end-to-end pipeline covering metadata filtering, archive validation, LaTeX extraction, text normalization, domain-aware tokenization, and dense transformer training under constrained compute (2xA100 GPUs). Through 24 experimental runs, we analyze training stability, scaling behavior, data yield losses, and infrastructure bottlenecks. Our findings highlight how preprocessing decisions significantly affect usable token volume, how tokenization impacts symbolic stability, and how storage and I/O constraints can rival compute as limiting factors. We further analyze convergence dynamics and show stable training behavior in a data-rich regime (52B pretraining tokens). Rather than proposing a novel architecture, this work provides an engineering-grounded, transparent account of training a small scientific language model from scratch. We hope these insights support researchers operating under moderate compute budgets who seek to build domain-specialized models.

Anuj Gupta

Independent Researcher India

Language Model Pretraining Language Modeling LLM Training

Feb 19, 2026

ALPS: A Diagnostic Challenge Set for Arabic Linguistic & Pragmatic Reasoning

While recent Arabic NLP benchmarks focus on scale, they often rely on synthetic or translated data which may benefit from deeper linguistic verification. We introduce ALPS (Arabic Linguistic & Pragmatic Suite), a native, expert-curated diagnostic challenge set probing Deep Semantics and Pragmatics, capabilities that complement specialized large-scale benchmarks. While broad-coverage benchmarks prioritize scale and multi-task coverage, ALPS targets the depth of linguistic understanding through 531 rigorously crafted questions across 15 tasks and 47 subtasks. We developed the dataset with deep expertise in Arabic linguistics, guaranteeing cultural authenticity and eliminating translation artifacts. Evaluating 23 diverse models (commercial, open-source, and Arabic-native) against a single-pass human performance (avg. 84.6% accuracy) and an expert-adjudicated oracle (99.2%), we reveal a critical dissociation: models achieve high fluency but fail on fundamental morpho-syntactic dependencies, with elevated error rates on morpho-syntactic dependencies (36.5% across diacritics-reliant tasks) compared to compositional semantics. While top commercial models (Gemini-3-flash at 94.2%) surpass the average single human, a substantial gap persists between commercial giants and Arabic-native models, with the best Arabic-specific model (Jais-2-70B at 83.6%) approaching but not matching human performance.

Hussein S. Al-Olimat, Ahmad Alshareef

Independent Researchers

Arabic NLP Arabic Language Models Pragmatic Reasoning in Language Models Language Modeling Multilingual Language Model Evaluation

Feb 18, 2026

Dialects of Translationese Shape Language Model Learning

Machine-translated data is widely used in multilingual NLP, particularly where native text is scarce. However, translated text differs systematically from native text. This phenomenon is known as translationese, and it reflects both traces of the source language and characteristic properties of translation itself. In this paper, we study how training on machine-translated data affects small English language models, focusing on how translationese from different source languages shapes linguistic acceptability judgments and language modeling for different domains. We train models on English text translated from 24 typologically and resource-diverse source languages, enabling a systematic analysis of how source language and corpus properties influence what models learn. Our results show that the source language has a clear impact on model behavior: general perplexity is more driven by the lexical diversity of the translated corpus, but grammatical performance is strongly correlated to typological similarity to English if trained on enough data. Even translation quality is a strong predictor of language modeling performance.

Jenny Kunz

Department of Computer and Information Science Linköping University

Language Modeling

Feb 17, 2026

Avey-B

Compact pretrained bidirectional encoders remain the backbone of industrial NLP under tight compute and memory budgets. Their effectiveness stems from self-attention's ability to deliver high-quality bidirectional contextualization with sequence-level parallelism, as popularized by BERT-style architectures. Recently, Avey was introduced as an autoregressive, attention-free alternative that naturally admits an encoder-only adaptation. In this paper, we reformulate Avey for the encoder-only paradigm and propose several innovations to its architecture, including decoupled static and dynamic parameterizations, stability-oriented normalization, and neural compression. Results show that this reformulated architecture compares favorably to four widely used Transformer-based encoders, consistently outperforming them on standard token-classification and information-retrieval benchmarks while scaling more efficiently to long contexts.

Devang Acharya, Mohammad Hammoud

Avey AI

Language Modeling Transformer Long-Context Modeling

Feb 16, 2026

RF-GPT: Teaching AI to See the Wireless World

Large language models (LLMs) and multimodal models have become powerful general-purpose reasoning systems. However, radio-frequency (RF) signals, which underpin wireless systems, are still not natively supported by these models. Existing LLM-based approaches for telecom focus mainly on text and structured data, while conventional RF deep-learning models are built separately for specific signal-processing tasks, highlighting a clear gap between RF perception and high-level reasoning. To bridge this gap, we introduce RF-GPT, a radio-frequency language model (RFLM) that utilizes the visual encoders of multimodal LLMs to process and understand RF spectrograms. In this framework, complex in-phase/quadrature (IQ) waveforms are mapped to time-frequency spectrograms and then passed to pretrained visual encoders. The resulting representations are injected as RF tokens into a decoder-only LLM, which generates RF-grounded answers, explanations, and structured outputs. To train RF-GPT, we perform supervised instruction fine-tuning of a pretrained multimodal LLM using a fully synthetic RF corpus. Standards-compliant waveform generators produce wideband scenes for six wireless technologies, from which we derive time-frequency spectrograms, exact configuration metadata, and dense captions. A text-only LLM then converts these captions into RF-grounded instruction-answer pairs, yielding roughly 120,000 RF scenes and 0.625 million instruction examples without any manual labeling. Across benchmarks for wideband modulation classification, overlap analysis, wireless-technology recognition, WLAN user counting, and 5G NR information extraction, RF-GPT achieves strong multi-task performance, whereas general-purpose VLMs with no RF grounding largely fail.

Hang Zou, Yu Tian, Bohao Wang +4

Research Institute for Digital Future, Khalifa University, 127788 Abu Dhabi, UAE · College of Information Science and Electronic Engineering, Zhejiang University, 310027, Hangzhou, China · Université de Lorraine, CNRS, CRAN, F-54000 Nancy, France

Wireless Communications Language Modeling Multimodal Large Language Models

Feb 16, 2026

Competition for attention predicts good-to-bad tipping in AI

More than half the global population now carries devices that can run ChatGPT-like language models with no Internet connection and minimal safety oversight -- and hence the potential to promote self-harm, financial losses and extremism among other dangers. Existing safety tools either require cloud connectivity or discover failures only after harm has occurred. Here we show that a large class of potentially dangerous tipping originates at the atomistic scale in such edge AI due to competition for the machinery's attention. This yields a mathematical formula for the dynamical tipping point n*, governed by dot-product competition for attention between the conversation's context and competing output basins, that reveals new control levers. Validated against multiple AI models, the mechanism can be instantiated for different definitions of 'good' and 'bad' and hence in principle applies across domains (e.g. health, law, finance, defense), changing legal landscapes (e.g. EU, UK, US and state level), languages, and cultural settings.

Neil F. Johnson, Frank Y. Huo

Physics Department and Dynamic Online Networks Laboratory, George Washington University, Washington D.C. 20052, U.S.A.

LLM Safety Language Modeling Attention Mechanisms

Feb 13, 2026

Left-right asymmetry in predicting brain activity from LLMs' representations emerges with their formal linguistic competence

When humans and large language models (LLMs) process the same text, activations in the LLMs correlate with brain activity measured, e.g., with functional magnetic resonance imaging (fMRI). Moreover, it has been shown that, as the training of an LLM progresses, the performance in predicting brain activity from its internal activations improves more in the left hemisphere than in the right one. The aim of the present work is to understand which kind of competence acquired by the LLMs underlies the emergence of this left-right asymmetry. Using the OLMo-2 7B language model at various training checkpoints and fMRI data from English participants, we compare the evolution of the left-right asymmetry in the correlation between brain activity and model predictions alongside performance on several benchmarks. We observe that the asymmetry co-emerges with the formal linguistic abilities of the LLM. These abilities are demonstrated in two ways: by the model's capacity to assign a higher probability to an acceptable sentence than to a grammatically unacceptable one within a minimal contrasting pair, and by its ability to produce well-formed text. By contrast, the left-right asymmetry does not align with the performance on arithmetic or Dyck language tasks; nor with text-based tasks involving world knowledge and reasoning. We generalize these results to another family of LLMs (Pythia) and two other languages, French and Chinese. Our observations indicate that the left-right asymmetry in brain predictivity matches the progress in formal linguistic competence.

Laurent Bonnasse-Gahot, Christophe Pallier

Centre d’Analyse et de Mathématique Sociales CNRS, EHESS, Paris, France · Cognitive Neuroimaging Unit CNRS, INSERM, CEA, Neurospin Center, 91191 Gif-sur-Yvette, France

Language Modeling

Feb 11, 2026

Compressed Sensing for Capability Localization in Large Language Models

Large language models (LLMs) exhibit a wide range of capabilities, including mathematical reasoning, code generation, and linguistic behaviors. We show that Transformer architectures contain small subsets of attention heads that are necessary for certain capabilities. Zeroing out as few as five task-specific heads can degrade performance by up to

60\%

on standard benchmarks measuring the capability of interest, while largely preserving performance on unrelated tasks. We introduce a compressed sensing-based method that exploits the sparsity of these heads to identify them via strategic knockouts and a small number of model evaluations. We validate these findings across Llama and Qwen models ranging from 1B to 14B parameters and a diverse set of capabilities including mathematical abilities and code generation, revealing a modular organization in which specialized capabilities are dependent on sparse, functionally distinct components. Overall, our results suggest that capability localization is a general organizational principle of Transformer language models, with implications for interpretability, model editing, and AI safety. Code is released at https://github.com/locuslab/llm-components.

Anna Bair, Yixuan Even Xu, Mingjie Sun +1

Carnegie Mellon University

Attention Head Analysis LLM Interpretability Language Modeling Compressed Sensing Attention Mechanisms

Feb 3, 2026

SalamahBench: Dialect and Category Level Safety Evaluation of Arabic Language Models

While different stakeholders are trying to leverage Arabic Language Models (ALMs), safety alignment in ALMs remains largely underexplored, hindering their mainstream adoption. Existing safety benchmarks are predominantly English-centric and evaluate Arabic only in its standardized form, obscuring fine-grained safety vulnerabilities in Arabic NLP systems. This paper introduces SalamahBench, a unified benchmark of 8{,}270 human-verified harmful prompts across ML Commons hazard categories, each rendered in Modern Standard Arabic (MSA) and five regional Arabic varieties, namely Egyptian, Syrian, Saudi, Lebanese, and Moroccan, for a total of 49{,}620 paired instances. To analyze the resulting data, we introduce two complementary metrics, namely Dialect Shift, which measures a model's aggregate change in safety under dialectal reformulation, and Category-Specific Dialect Deviation, which isolates harm categories whose change departs from that aggregate trend. Evaluating models such as Fanar 2, ALLaM 2, and Karnak 1 under multiple safeguard configurations, we find that cross-variety robustness is strongly model dependent, and that aggregate scores can conceal category-level divergence. Our findings highlight the necessity of evaluating Arabic model safety jointly across linguistic varieties and harm domains rather than relying on aggregate scores or MSA alone.

Omar Abdelnasser, Fatemah Alharbi, Khaled Khasawneh +2

Compumacy for Artificial Intelligence Solutions, Cairo, Egypt · Computer Science Department, College of Computer Science and Engineering, Taibah University, Yanbu 46522, KSA · Electrical and Computer Engineering Department, George Mason University, VA, USA +2

LLM Safety Benchmarks Multilingual Language Model Evaluation Arabic NLP Arabic Language Models Language Model Safety Evaluation Language Modeling+2

Feb 2, 2026

From Directions to Regions: Decomposing Activations in Language Models via Local Geometry

Activation decomposition methods in language models are tightly coupled to geometric assumptions on how concepts are realized in activation space. Existing approaches search for individual global directions, implicitly assuming linear separability, which overlooks concepts with nonlinear or multi-dimensional structure. In this work, we leverage Mixture of Factor Analyzers (MFA) as a scalable, unsupervised alternative that models the activation space as a collection of Gaussian regions with their local covariance structure. MFA decomposes activations into two compositional geometric objects: the region's centroid in activation space, and the local variation from the centroid. We train large-scale MFAs for Llama-3.1-8B and Gemma-2-2B, and show they capture complex, nonlinear structures in activation space. Moreover, evaluations on localization and steering benchmarks show that MFA outperforms unsupervised baselines, is competitive with supervised localization methods, and often achieves stronger steering performance than sparse autoencoders. Together, our findings position local geometry, expressed through subspaces, as a promising unit of analysis for scalable concept discovery and model control, accounting for complex structures that isolated directions fail to capture.

Or Shafran, Shaked Ronen, Omri Fahn +3

Blavatnik School of Computer Science and AI, Tel Aviv University, Israel · New York University, New York, NY, USA · Goodfire.

Language Model Steering LLM Interpretability Gaussian Mixture Models Language Modeling Representation Geometry in Language Models Concept Localization in Language Models

Feb 1, 2026

Distilling Token-Trained Models into Byte-Level Models

Byte Language Models (BLMs) have emerged as a promising direction for scaling language models beyond tokenization. However, existing BLMs typically require training from scratch on trillions of bytes, making them prohibitively expensive. In this paper, we propose an efficient distillation recipe that converts existing token-trained LLMs into BLMs while retaining comparable capabilities. Our recipe follows a two-stage curriculum: (1) Progressive Knowledge Distillation, which aligns byte-level representations with the embeddings of the token-trained teacher model; and (2) Byte-Level Supervised Fine-Tuning, which enables end-to-end generation entirely in the byte space. We validate our approach across multiple model families, including Llama, Qwen, and OLMo, and demonstrate that the distilled BLMs retain most of the teacher models' performance using only approximately 125B bytes.

Zishuo Bao, Jiaqi Leng, Junxiong Wang +2

Fuzhou University · NYU Shanghai · Fudan University +2

LLM Fine-Tuning Byte-Level Language Model Language Model Distillation Language Modeling

Jan 27, 2026

Understanding LLM Failures: A Multi-Tape Turing Machine Analysis of Systematic Errors in Language Model Reasoning

Large language models (LLMs) exhibit failure modes on seemingly trivial tasks. We propose a formalisation of LLM interaction using a deterministic multi-tape Turing machine, where each tape represents a distinct component: input characters, tokens, vocabulary, model parameters, activations, probability distributions, and output text. The model enables precise localisation of failure modes to specific pipeline stages, revealing, e.g., how tokenisation obscures character-level structure needed for counting tasks. The model clarifies why techniques like chain-of-thought prompting help, by externalising computation on the output tape, while also revealing their fundamental limitations. This approach provides a rigorous, falsifiable alternative to geometric metaphors and complements empirical scaling laws with principled error analysis.

Magnus Boman

Karolinska Institutet, Stockholm, Sweden

Tokenization in Language Models LLM Interpretability Language Modeling LLM Reasoning

Jan 21, 2026

Mechanism Shift During Post-training from Autoregressive to Masked Diffusion Language Models

Post-training pretrained autoregressive models (ARMs) into masked diffusion models (MDMs) provides an efficient route to diffusion language modeling, but it remains unclear whether the resulting models reuse inherited autoregressive computation or reorganize it for non-autoregressive generation. We compare two 7B ARM-MDM families across four controlled diagnostic tasks and find a task-dependent mechanism shift. On prefix-dominant tasks, MDMs largely preserve inherited high-attribution pathways or exhibit only modest changes in where computation occurs. On globally constrained tasks, the reorganization is substantially stronger, with task-relevant computation shifting toward earlier layers. This depth-wise pattern persists across prompt resampling, circuit budgets, and tested inference budgets, while targeted ablations support the functional importance of the identified structures under the tested intervention protocols. At the component level, diagnostic probes suggest that ARMs rely more strongly on sharply specialized components, whereas MDMs exhibit weaker single-component specialization and more diffuse output-space alignment. Together, these results suggest that diffusion post-training selectively preserves or reorganizes inherited computation according to task structure, rather than uniformly replacing autoregressive mechanisms.

Injin Kong, Hyoungjoon Lee, Yohan Jo

Graduate School of Data Science, Seoul National University · Department of Biosystems & Biomaterials Science and Engineering, Seoul National University

Masked Diffusion Models Autoregressive Diffusion Autoregressive Language Modeling Mechanistic Interpretability Diffusion Model Fine-Tuning Language Modeling+2

Jan 12, 2026

Conditional Memory via Scalable Lookup: A New Axis of Sparsity for Large Language Models

While Mixture-of-Experts (MoE) scales capacity via conditional computation, Transformers lack a native primitive for knowledge lookup, forcing them to inefficiently simulate retrieval through computation. To address this, we introduce conditional memory as a complementary sparsity axis, instantiated via Engram, a module that modernizes classic

N

-gram embedding for O(1) lookup. By formulating the Sparsity Allocation problem, we uncover a U-shaped scaling law that optimizes the trade-off between neural computation (MoE) and static memory (Engram). Guided by this law, we scale Engram to 27B parameters, achieving superior performance over a strictly iso-parameter and iso-FLOPs MoE baseline. Most notably, while the memory module is expected to aid knowledge retrieval (e.g., MMLU +3.4; CMMLU +4.0), we observe even larger gains in general reasoning (e.g., BBH +5.0; ARC-Challenge +3.7) and code/math domains~(HumanEval +3.0; MATH +2.4). Mechanistic analyses reveal that Engram relieves the backbone's early layers from static reconstruction, effectively deepening the network for complex reasoning. Furthermore, by delegating local dependencies to lookups, it frees up attention capacity for global context, substantially boosting long-context retrieval (e.g., Multi-Query NIAH: 84.2 to 97.0). Finally, Engram establishes infrastructure-aware efficiency: its deterministic addressing enables runtime prefetching from host memory, incurring negligible overhead. We envision conditional memory as an indispensable modeling primitive for next-generation sparse models.

Xin Cheng, Rui Tian, Wangding Zeng +18

Peking University · DeepSeek-AI

Memory-Augmented Language Models Language Model Scaling Laws Language Modeling