Committee

arXiv:2512.06921v1 Announce Type: cross Abstract: Multimodal Large Language Models (MLLMs) have shown significant potential in surgical video understanding. With improved zero-shot performance and more effective human-machine interaction, they provide a strong foundation for advancing surgical education and assistance. However, existing research and datasets primarily focus on understanding surgical procedures and workflows, while paying limited attention to the critical role of anatomical comprehension. In clinical practice, surgeons rely heavily on precise anatomical understanding to interpret, review, and learn from surgical videos. To fill this gap, we introduce the Neurosurgical Anatomy Benchmark (NeuroABench), the first multimodal benchmark explicitly created to evaluate anatomical understanding in the neurosurgical domain. NeuroABench consists of 9 hours of annotated neurosurgical videos covering 89 distinct procedures and is developed using a novel multimodal annotation pipeline with multiple review cycles. The benchmark evaluates the identification of 68 clinical anatomical structures, providing a rigorous and standardized framework for assessing model performance. Experiments on over 10 state-of-the-art MLLMs reveal significant limitations, with the best-performing model achieving only 40.87% accuracy in anatomical identification tasks. To further evaluate the benchmark, we extract a subset of the dataset and conduct an informative test with four neurosurgical trainees. The results show that the best-performing student achieves 56% accuracy, with the lowest scores of 28% and an average score of 46.5%. While the best MLLM performs comparably to the lowest-scoring student, it still lags significantly behind the group’s average performance. This comparison underscores both the progress of MLLMs in anatomical understanding and the substantial gap that remains in achieving human-level performance.

arXiv:2512.06922v1 Announce Type: new Abstract: Large language models (LLMs) present a dual challenge for forensic linguistics. They serve as powerful analytical tools enabling scalable corpus analysis and embedding-based authorship attribution, while simultaneously destabilising foundational assumptions about idiolect through style mimicry, authorship obfuscation, and the proliferation of synthetic texts. Recent stylometric research indicates that LLMs can approximate surface stylistic features yet exhibit detectable differences from human writers, a tension with significant forensic implications. However, current AI-text detection techniques, whether classifier-based, stylometric, or watermarking approaches, face substantial limitations: high false positive rates for non-native English writers and vulnerability to adversarial strategies such as homoglyph substitution. These uncertainties raise concerns under legal admissibility standards, particularly the Daubert and Kumho Tire frameworks. The article concludes that forensic linguistics requires methodological reconfiguration to remain scientifically credible and legally admissible. Proposed adaptations include hybrid human-AI workflows, explainable detection paradigms beyond binary classification, and validation regimes measuring error and bias across diverse populations. The discipline’s core insight, i.e., that language reveals information about its producer, remains valid but must accommodate increasingly complex chains of human and machine authorship.

arXiv:2512.06350v1 Announce Type: cross Abstract: The emergence of transformative technologies often surfaces deep societal divisions, nowhere more evident than in contemporary debates about artificial intelligence (AI). A striking feature of these divisions is that they persist despite shared interests in ensuring that AI benefits humanity and avoiding catastrophic outcomes. This paper analyzes contemporary debates about AI risk, parsing the differences between the “doomer” and “boomer” perspectives into definitional, factual, causal, and moral premises to identify key points of contention. We find that differences in perspectives about existential risk (“X-risk”) arise fundamentally from differences in causal premises about design vs. emergence in complex systems, while differences in perspectives about employment risks (“E-risks”) pertain to different causal premises about the applicability of past theories (evolution) vs their inapplicability (revolution). Disagreements about these two forms of AI risk appear to share two properties: neither involves significant disagreements on moral values and both can be described in terms of differing views on the extent of boundedness of human rationality. Our approach to analyzing reasoning chains at scale, using an ensemble of LLMs to parse textual data, can be applied to identify key points of contention in debates about risk to the public in any arena.

arXiv:2512.05700v1 Announce Type: new Abstract: We present a methodology for improving the accuracy of faithfulness evaluation in Large Language Models (LLMs). The proposed methodology is based on the combination of elementary faithfulness metrics into a combined (fused) metric, for the purpose of improving the faithfulness of LLM outputs. The proposed strategy for metric fusion deploys a tree-based model to identify the importance of each metric, which is driven by the integration of human judgements evaluating the faithfulness of LLM responses. This fused metric is demonstrated to correlate more strongly with human judgements across all tested domains for faithfulness. Improving the ability to evaluate the faithfulness of LLMs, allows for greater confidence to be placed within models, allowing for their implementation in a greater diversity of scenarios. Additionally, we homogenise a collection of datasets across question answering and dialogue-based domains and implement human judgements and LLM responses within this dataset, allowing for the reproduction and trialling of faithfulness evaluation across domains.

arXiv:2510.04996v3 Announce Type: replace-cross Abstract: Reinforcement learning (RL) for large language model reasoning is frequently hindered by signal loss, a phenomenon where standard uniform sampling with small group sizes fails to uncover informative learning signals for difficult prompts. We demonstrate that this collapse is a statistical artifact of undersampling rather than an inherent model limitation. To address this systematically, we introduce a theoretical framework based on optimizing a non-linear RL objective (e.g., log-likelihood). We show that this objective naturally induces a weighted gradient estimator that prioritizes difficult prompts, which can be robustly realized through adaptive sampling. Guided by this framework, we propose Reinforce-Ada, a family of algorithms that dynamically allocates inference budgets based on prompt difficulty, effectively scaling up RL compute to where it is needed most. Unlike passive filtering methods that discard low-signal prompts, Reinforce-Ada actively invests compute to recover them. We introduce two efficient realizations: an estimation-based approach and a model-free sequential sampling approach. Extensive experiments across multiple benchmarks show that Reinforce-Ada significantly outperforms uniform baselines like GRPO, recovering lost signals and accelerating convergence by up to $2times$ while maintaining the same total inference budget. Code is available at https://github.com/RLHFlow/Reinforce-Ada.

arXiv:2512.05373v1 Announce Type: cross Abstract: Recent advances in natural language processing have enabled the increasing use of text data in causal inference, particularly for adjusting confounding factors in treatment effect estimation. Although high-dimensional text can encode rich contextual information, it also poses unique challenges for causal identification and estimation. In particular, the positivity assumption, which requires sufficient treatment overlap across confounder values, is often violated at the observational level, when massive text is represented in feature spaces. Redundant or spurious textual features inflate dimensionality, producing extreme propensity scores, unstable weights, and inflated variance in effect estimates. We address these challenges with Confounding-Aware Token Rationalization (CATR), a framework that selects a sparse necessary subset of tokens using a residual-independence diagnostic designed to preserve confounding information sufficient for unconfoundedness. By discarding irrelevant texts while retaining key signals, CATR mitigates observational-level positivity violations and stabilizes downstream causal effect estimators. Experiments on synthetic data and a real-world study using the MIMIC-III database demonstrate that CATR yields more accurate, stable, and interpretable causal effect estimates than existing baselines.

arXiv:2512.05387v1 Announce Type: new Abstract: Large Language Models (LLMs) often suffer from hallucinations: output content that is not grounded in the input context, when performing long-form text generation tasks such as summarization. Prior works have shown that hallucinations can be reduced by iteratively critiquing and refining previously generated outputs using either the same model or a more powerful teacher model as the critique. However, these approaches either require additional test-time compute or assume access to more powerful teacher models, making them costly and less practical. In this work, we propose Self Critique and Refinement-based Preference Optimization (SCRPO), which is a self-supervised training framework that first constructs a preference dataset by leveraging the LLM’s own critique and refinement capabilities, and then applies preference learning to improve the same LLM for faithful summarization. Experiments on three summarization benchmarks (XSUM CNNDM and SAMSum), demonstrate that our approach outperforms state-of-the-art self-supervised learning methods in terms of faithfulness metrics while either maintaining or improving other metrics that measure the overall quality of the summary. Moreover, compared to test-time refinement, our approach not only improves efficiency but also results in more faithful summaries.

arXiv:2503.07691v2 Announce Type: replace-cross Abstract: Group fairness is a central research topic in text classification, where reaching fair treatment between sensitive groups (e.g., women and men) remains an open challenge. We propose an approach that extends the use of the Wasserstein Dependency Measure for learning unbiased neural text classifiers. Given the challenge of distinguishing fair from unfair information in a text encoder, we draw inspiration from adversarial training by inducing independence between representations learned for the target label and those for a sensitive attribute. We further show that Domain Adaptation can be efficiently leveraged to remove the need for access to the sensitive attributes in the dataset we cure. We provide both theoretical and empirical evidence that our approach is well-founded.

Microsoft has released VibeVoice-Realtime-0.5B, a real time text to speech model that works with streaming text input and long form speech output, aimed at agent style applications and live data narration. The model can start producing audible speech in about 300 ms, which is critical when a language model is still generating the rest of its answer. Where VibeVoice Realtime Fits in the VibeVoice Stack? VibeVoice is a broader framework that focuses on next token diffusion over continuous speech tokens, with variants designed for long form multi speaker audio such as podcasts. The research team shows that the main VibeVoice models can synthesize up to 90 minutes of speech with up to 4 speakers in a 64k context window using continuous speech tokenizers at 7.5 Hz. The Realtime 0.5B variant is the low latency branch of this family. The model card reports an 8k context length and a typical generation length of about 10 minutes for a single speaker, which is enough for most voice agents, system narrators and live dashboards. A separate set of VibeVoice models, VibeVoice-1.5B and VibeVoice Large, handle long form multi speaker audio with 32k and 64k context windows and longer generation times. Interleaved Streaming Architecture The realtime variant uses an interleaved windowed design. Incoming text is split into chunks. The model incrementally encodes new text chunks while, in parallel, continuing diffusion based acoustic latent generation from prior context. This overlap between text encoding and acoustic decoding is what lets the system reach about 300 ms first audio latency on suitable hardware. Unlike the long form VibeVoice variants, which use both semantic and acoustic tokenizers, the realtime model removes the semantic tokenizer and uses only an acoustic tokenizer that operates at 7.5 Hz. The acoustic tokenizer is based on a σ VAE variant from LatentLM, with a mirror symmetric encoder decoder architecture that uses 7 stages of modified transformer blocks and performs 3200x downsampling from 24 kHz audio. On top of this tokenizer, a diffusion head predicts acoustic VAE features. The diffusion head has 4 layers and about 40M parameters and is conditioned on hidden states from Qwen2.5-0.5B. It uses a Denoising Diffusion Probabilistic Models process with Classifier Free Guidance and DPM Solver style samplers, following the next token diffusion approach of the full VibeVoice system. Training proceeds in two stages. First, the acoustic tokenizer is pre trained. Then the tokenizer is frozen and the team trains the LLM along with the diffusion head with curriculum learning on sequence length, increasing from about 4k to 8,192 tokens. This keeps the tokenizer stable, while the LLM and diffusion head learn to map from text tokens to acoustic tokens across long contexts. Quality on LibriSpeech and SEED The VibeVoice Realtime reports zero shot performance on LibriSpeech test clean. VibeVoice Realtime 0.5B reaches word error rate (WER) 2.00 percent and speaker similarity 0.695. For comparison, VALL-E 2 has WER 2.40 with similarity 0.643 and Voicebox has WER 1.90 with similarity 0.662 on the same benchmark. On the SEED test benchmark for short utterances, VibeVoice Realtime-0.5B reaches WER 2.05 percent and speaker similarity 0.633. SparkTTS gets a slightly lower WER 1.98 but lower similarity 0.584, while Seed TTS reaches WER 2.25 and the highest reported similarity 0.762. The research team noted that the realtime model is optimized for long form robustness, so short sentence metrics are informative but not the main target. From an engineering point of view, the interesting part is the tradeoff. By running the acoustic tokenizer at 7.5 Hz and using next token diffusion, the model reduces the number of steps per second of audio compared to higher frame rate tokenizers, while preserving competitive WER and speaker similarity. Integration Pattern for Agents And Applications The recommended setup is to run VibeVoice-Realtime-0.5B next to a conversational LLM. The LLM streams tokens during generation. These text chunks feed directly into the VibeVoice server, which synthesizes audio in parallel and streams it back to the client. For many systems this looks like a small microservice. The TTS process has a fixed 8k context and about 10 minutes of audio budget per request, which fits typical agent dialogs, support calls and monitoring dashboards. Because the model is speech only and does not generate background ambience or music, it is better suited for voice interfaces, assistant style products and programmatic narration rather than media production. Key Takeaways Low latency streaming TTS: VibeVoice-Realtime-0.5B is a real time text to speech model that supports streaming text input and can emit the first audio frames in about 300 ms, which makes it suitable for interactive agents and live narration where users cannot tolerate 1 to 3 second delays. LLM along with diffusion over continuous speech tokens: The model follows the VibeVoice design, it uses a Qwen2.5 0.5B language model to process text context and dialogue flow, then a diffusion head operates on continuous acoustic tokens from a low frame rate tokenizer to generate waveform level detail, which scales better to long sequences than classic spectrogram based TTS. Around 1B total parameters with acoustic stack: While the base LLM has 0.5B parameters, the acoustic decoder has about 340M parameters and the diffusion head about 40M parameters, so the full realtime stack is roughly 1B parameters, which is important for GPU memory planning and deployment sizing. Competitive quality on LibriSpeech and SEED: On LibriSpeech test clean, VibeVoice-Realtime-0.5B reaches word error rate 2.00 percent and speaker similarity 0.695, and on SEED test en it reaches 2.05 percent WER and 0.633 similarity, which places it in the same quality band as strong recent TTS systems while still being tuned for long form robustness. Check out the Model Card on HF. Feel free to check out our GitHub Page for Tutorials, Codes and Notebooks. Also, feel free to follow us on Twitter and don’t forget to join our 100k+ ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well. The post Microsoft AI Releases VibeVoice-Realtime: A Lightweight Real‑Time Text-to-Speech Model Supporting Streaming Text Input and Robust Long-Form Speech Generation appeared first on MarkTechPost.