arXiv subjects

Vatsal Baherwani

Publications and source records attributed to Vatsal Baherwani.

5 recordsLinked to original sources

arXiv2026

Not All LLM Reasoning is Visible in the Chain-of-Thought

A key question for AI safety is whether a language model expresses all of its reasoning in its output tokens. We demonstrate a concrete failure mode where frontier models exhibit invisible reasoning by leveraging semantically irrelevant filler tokens to improve performance on synthetic reasoning tasks. We evaluate 13 frontier language models across three tasks and find that many models benefit significantly from filler tokens, with accuracy improvements of up to 13 percentage points. The benefit depends on which tokens are used and differs across models. We further show that filler tokens enable Claude Opus 4.5 to satisfy a hidden modular arithmetic constraint without sacrificing accuracy on its primary task, demonstrating that invisible reasoning can serve objectives entirely invisible to CoT monitoring. Reinforcement learning gives Qwen3-235B strong preferences over filler token content, but neither RL nor supervised fine-tuning produces a filler token benefit that persists at test time. Our results indicate that frontier models already perform consequential computation with no interpretable trace in their output tokens.

cs.CL↗

arXiv2026

Emergent Capabilities Arise Randomly from Learning Sparse Attention Patterns

Neural scaling laws for transformer language models predict smooth improvements in pretraining loss with increasing parameters, but downstream capabilities such as in-context learning are known to emerge abruptly past a certain model scale. In this paper, we show that emergent capabilities arise stochastically throughout training, with larger models acquiring them earlier on average. We demonstrate that the emergence of capabilities such as pattern completion and indirect object identification corresponds to the abrupt learning of task-relevant attention patterns. To isolate this phenomenon, we train transformer models on synthetic linear map and cellular automata datasets, and we show that the difficulty of learning attention patterns depends on context length and pattern sparsity. Moreover, scaling the number of attention heads improves learning efficiency on our synthetic tasks, while increasing the head dimension yields diminishing returns past a minimum capacity. We additionally investigate architectures with alternative attention mechanisms, showing that MLP-Mixer outperforms a transformer on linear map tasks with complex attention patterns. Our findings provide a mechanistic insight into emergence, showing that downstream capabilities arise abruptly due to the intrinsic difficulty of learning sparse attention patterns in transformer models.

cs.LG↗

arXiv2025

Characterizing Motion Encoding in Video Diffusion Timesteps

Text-to-video diffusion models synthesize temporal motion and spatial appearance through iterative denoising, yet how motion is encoded across timesteps remains poorly understood. Practitioners often exploit the empirical heuristic that early timesteps mainly shape motion and layout while later ones refine appearance, but this behavior has not been systematically characterized. In this work, we proxy motion encoding in video diffusion timesteps by the trade-off between appearance editing and motion preservation induced when injecting new conditions over specified timestep ranges, and characterize this proxy through a large-scale quantitative study. This protocol allows us to factor motion from appearance by quantitatively mapping how they compete along the denoising trajectory. Across diverse architectures, we consistently identify an early, motion-dominant regime and a later, appearance-dominant regime, yielding an operational motion-appearance boundary in timestep space. Building on this characterization, we simplify current one-shot motion customization paradigm by restricting training and inference to the motion-dominant regime, achieving strong motion transfer without auxiliary debiasing modules or specialized objectives. Our analysis turns a widely used heuristic into a spatiotemporal disentanglement principle, and our timestep-constrained recipe can serve as ready integration into existing motion transfer and editing methods.

cs.CV↗

arXiv2025

Dense Backpropagation Improves Training for Sparse Mixture-of-Experts

Mixture of Experts (MoE) pretraining is more scalable than dense Transformer pretraining, because MoEs learn to route inputs to a sparse set of their feedforward parameters. However, this means that MoEs only receive a sparse backward update, leading to training instability and suboptimal performance. We present a lightweight approximation method that gives the MoE router a dense gradient update while continuing to sparsely activate its parameters. Our method, which we refer to as Default MoE, substitutes missing expert activations with default outputs consisting of an exponential moving average of expert outputs previously seen over the course of training. This allows the router to receive signals from every expert for each token, leading to significant improvements in training performance. Our Default MoE outperforms standard TopK routing in a variety of settings without requiring significant computational overhead. Code: https://github.com/vatsal0/default-moe.

cs.LG↗

arXiv2025

DynaGuard: A Dynamic Guardian Model With User-Defined Policies

Guardian models play a crucial role in ensuring the safety and ethical behavior of user-facing AI applications by enforcing guardrails and detecting harmful content. While standard guardian models are limited to predefined, static harm categories, we introduce DynaGuard, a suite of dynamic guardian models offering novel flexibility by evaluating text based on user-defined policies, and DynaBench, a dataset for training and evaluating dynamic guardian models. Our models provide both rapid detection of policy violations and a chain-of-thought reasoning option that articulate and justify model outputs. Critically, DynaGuard not only surpasses static models in detection accuracy on traditional safety categories, but is competitive with frontier reasoning models on free-form policy violations, all in a fraction of the time. This makes DynaGuard an critical tool for language model guardrails.

cs.LG↗