Diffusion language models, especially masked discrete diffusion models, have achieved great success recently. While there are some theoretical and primary empirical results showing the advantages of latent reasoning with looped transformers or continuous chain-of-thoughts, continuous diffusion models typically underperform their discrete counterparts. In this paper, we argue that diffusion language models do not necessarily need to be in the discrete space. In particular, we prove that continuous diffusion models have stronger expressivity than discrete diffusions and looped transformers. We attribute the contradiction between the theoretical expressiveness and empirical performance to their practical trainability: while continuous diffusion provides intermediate supervision that looped transformers lack, they introduce additional difficulty decoding tokens into the discrete token space from the continuous representation space. We therefore propose Coevolutionary Continuous Discrete Diffusion (CCDD), which defines a joint multimodal diffusion process on the union of a continuous representation space and a discrete token space, leveraging a single model to simultaneously denoise in the joint space. By combining two modalities, CCDD is expressive with rich semantics in the latent space, as well as good trainability and sample quality with the help of explicit discrete tokens. We also propose effective architectures and advanced training/sampling techniques for CCDD, which reveals strong empirical performance in extensive language modeling experiments on real-world tasks.

Large Language Models (LLMs) achieve strong results on code tasks, but how they derive program meaning remains unclear. We argue that code communicates through two channels: structural semantics, which define formal behavior, and human-interpretable naming, which conveys intent. Surprisingly, we also observe consistent reductions on execution tasks that should depend only on structure, revealing that current benchmarks reward memorization of naming patterns rather than genuine semantic reasoning. To disentangle these effects, we introduce a suite of semantics-preserving obfuscations and show that they expose identifier leakage across both summarization and execution. Large language models (LLMs) now achieve striking results across code intelligence--program synthesis, repair, summarization, and test generation. Y et how these models derive meaning from source code remains unclear. If an LLM truly understands a program's intent, its behavior should remain stable when human-interpretable names are perturbed while semantics stay fixed; conversely, strong performance drops would indicate an overreliance on surface cues rather than semantic reasoning.

A significant proportion of queries to large language models ask them to edit user-provided text, rather than generate new text from scratch. While previous work focuses on detecting fully AI-generated text, we demonstrate that AI-edited text is distinguishable from human-written and AI-generated text. First, we propose using lightweight similarity metrics to quantify the magnitude of AI editing present in a text given the original human-written text and validate these metrics with human annotators. Using these similarity metrics as intermediate supervision, we then train EditLens, a regression model that predicts the amount of AI editing present within a text. Our model achieves state-of-the-art performance on both binary (F1=94.7%) and ternary (F1=90.4%) classification tasks in distinguishing human, AI, and mixed writing. Not only do we show that AI-edited text can be detected, but also that the degree of change made by AI to human writing can be detected, which has implications for authorship attribution, education, and policy. Finally, as a case study, we use our model to analyze the effects of AI-edits applied by Grammarly, a popular writing assistance tool. To encourage further research, we commit to publicly releasing our models and dataset.

Test-time algorithms that combine the generative power of language models with process verifiers that assess the quality of partial generations offer a promising lever for eliciting new reasoning capabilities, but the algorithmic design space and computational scaling properties of such approaches are still opaque, and their benefits are far from apparent when one accounts for the cost of learning a high-quality verifier. Our starting point is the observation that seemingly benign errors in a learned verifier can lead to catastrophic failures for standard decoding techniques due to error amplification during the course of generation. We then ask: can this be improved with more sophisticated decoding strategies? We introduce a new process-guided test-time sampling algorithm, VGB, which uses theoretically grounded backtracking to achieve provably better robustness to verifier errors. VGB interprets autoregressive generation as a random walk on a tree of partial generations, with transition probabilities guided by the process verifier and base model; crucially, backtracking occurs probabilistically. This process generalizes the seminal Sinclair-Jerrum random walk (Sinclair & Jerrum, 1989) from the literature on approximate counting and sampling in theoretical computer science, and a conceptual contribution of our work is to highlight parallels with this literature. Empirically, we demonstrate on both synthetic and real language modeling tasks that VGB outperforms baselines on a variety of metrics.

Confidence calibration, the alignment of a model's predicted confidence with its actual accuracy, is crucial for the reliable deployment of Large Language Models (LLMs). However, this critical property remains largely under-explored in multilingual contexts. In this work, we conduct the first large-scale, systematic studies of multilingual calibration across six model families and over 100 languages, revealing that non-English languages suffer from systematically worse calibration. To diagnose this, we investigate the model's internal representations and find that the final layer, biased by English-centric training, provides a poor signal for multilingual confidence. In contrast, our layer-wise analysis uncovers a key insight that late-intermediate layers consistently offer a more reliable and better-calibrated signal. Building on this, we introduce a suite of training-free methods, including Language-Aware Confidence Ensemble (LACE), which adaptively selects an optimal ensemble of layers for each specific language. Our study highlights the hidden costs of English-centric alignment and offer a new path toward building more globally equitable and trustworthy LLMs by looking beyond the final layer.

Analogical reasoning lies at the core of human cognition and remains a fundamental challenge for artificial intelligence. Raven's Progressive Matrices (RPM) serve as a widely used benchmark to assess abstract reasoning by requiring the inference of underlying structural rules. While many vision-based and language-based models have achieved success on RPM tasks, it remains unclear whether their performance reflects genuine reasoning ability or reliance on statistical shortcuts. This study investigates the generalization capacity of modern AI systems under conditions of incomplete training by deliberately omitting several structural rules during training. Both sequence-to-sequence transformer models and vision-based architectures such as CoPINet and the Dual-Contrast Network are evaluated on the Impartial-RAVEN (I-RAVEN) dataset. Experiments reveal that although transformers demonstrate strong performance on familiar rules, their accuracy declines sharply when faced with novel or omitted rules. Moreover, the gap between token-level accuracy and complete answer accuracy highlights fundamental limitations in current approaches. These findings provide new insights into the reasoning mechanisms underlying deep learning models and underscore the need for architectures that move beyond pattern recognition toward robust abstract reasoning.

Efficient real-time dispatching in urban metro systems is essential for ensuring service reliability, maximizing resource utilization, and improving passenger satisfaction. This study presents a novel deep learning framework centered on a Convolutional Long Short-Term Memory (ConvLSTM) model designed to predict the complex spatiotemporal propagation of train headways across an entire metro line. By directly incorporating planned terminal headways as a critical input alongside historical headway data, the proposed model accurately forecasts future headway dynamics, effectively capturing both their temporal evolution and spatial dependencies across all stations. This capability empowers dispatchers to evaluate the impact of various terminal headway control decisions without resorting to computationally intensive simulations. We introduce a flexible methodology to simulate diverse dispatcher strategies, ranging from maintaining even headways to implementing custom patterns derived from observed terminal departures. In contrast to existing research primarily focused on passenger load predictioning or atypical disruption scenarios, our approach emphasizes proactive operational control. Evaluated on a large-scale dataset from an urban metro line, the proposed ConvLSTM model demonstrates promising headway predictions, offering actionable insights for real-time decision-making. This framework provides rail operators with a powerful, computationally efficient tool to optimize dispatching strategies, thereby significantly improving service consistency and passenger satisfaction.

Speech-to-Text Translation (S2TT) systems built from Automatic Speech Recognition (ASR) and Text-to-Text Translation (T2TT) modules face two major limitations: error propagation and the inability to exploit prosodic or other acoustic cues. Chain-of-Thought (CoT) prompting has recently been introduced, with the expectation that jointly accessing speech and transcription will overcome these issues. Analyzing CoT through attribution methods, robustness evaluations with corrupted transcripts, and prosody-awareness, we find that it largely mirrors cascaded behavior, relying mainly on transcripts while barely leveraging speech. Simple training interventions, such as adding Direct S2TT data or noisy transcript injection, enhance robustness and increase speech attribution. These findings challenge the assumed advantages of CoT and highlight the need for architectures that explicitly integrate acoustic information into translation.

To utilize pre-trained neural networks on edge and mobile devices, we often require efficient adaptation to user-specific runtime data distributions while operating under limited compute and memory resources. On-device retraining with a target dataset can facilitate such adaptations; however, it remains impractical due to the increasing depth of modern neural nets, as well as the computational overhead associated with gradient-based optimization across all layers. Current approaches reduce training cost by selecting a subset of layers for retraining, however, they rely on labeled data, at least one full-model backpropagation, or server-side meta-training; limiting their suitability for constrained devices. We introduce AdaBet, a gradient-free layer selection approach to rank important layers by analyzing topological features of their activation spaces through Betti Numbers and using forward passes alone. AdaBet allows selecting layers with high learning capacity, which are important for retraining and adaptation, without requiring labels or gradients. Evaluating AdaBet on sixteen pairs of benchmark models and datasets, shows AdaBet achieves an average gain of 5% more classification accuracy over gradient-based baselines while reducing average peak memory consumption by 40%.

Speech Emotion Recognition (SER) is essential for improving human-computer interaction, yet its accuracy remains constrained by the complexity of emotional nuances in speech. In this study, we distinguish between descriptive semantics, which represents the contextual content of speech, and expressive semantics, which reflects the speaker's emotional state. After watching emotionally charged movie segments, we recorded audio clips of participants describing their experiences, along with the intended emotion tags for each clip, participants' self-rated emotional responses, and their valence/arousal scores. Through experiments, we show that descriptive semantics align with intended emotions, while expressive semantics correlate with evoked emotions. Our findings inform SER applications in human-AI interaction and pave the way for more context-aware AI systems.