AI research papers.

Rubric-based reinforcement learning (Rubric-RL) trains language models where no verifier exists.

Length scaling during reinforcement-learning (RL) post-training is often viewed as a sign of improved reasoning ability, especially on difficult problems, but may also make responses to already-solved problems unnecessar…

Large language models rely heavily on human text, which often conveys surface answers rather than the spatial and structural logic behind them.

Recent advances in robot learning have enabled manipulation policies to perform increasingly diverse tasks and generalize across environments.

Working agents need to read diverse files, coordinate tools, and produce deliverables.

Chain-of-thought (CoT) traces often serve as a proxy for how Large Language Models (LLMs) arrive at their answers.

Preference distillation typically treats a teacher response as preferred and the student's own response as rejected.

AI reviewers can assign different judgments to manuscripts that report the same science in different wording, potentially rewarding rhetorical optimization over scientific improvement.

Recent advances in large language models have made automatic game generation increasingly feasible, yet reliably improving generated games beyond a playable version remains challenging.

As multi-agent systems enter high-stakes domains, the possibility that agents may circumvent safety boundaries is a growing concern.

Long-context LLM agents accumulate interaction histories that strain KV-cache memory and attention computation.

Legal language models require reward signals that capture not only answer correctness but also the multidimensional quality of legal responses.

Self-evolving search agents build their own training curricula by jointly optimizing a proposer that generates questions and a solver that answers them.

Deep research tasks require agents to navigate large knowledge spaces, synthesize evidence across many sources, and adapt their plans as findings emerge.

Persistent spatial memory enables embodied agents to navigate familiar environments across repeated visits.

High-compression tokenizers are essential for scaling latent image generative models.

Iterative self-distillation enables LLM agents to learn from successive deployments, offering a path toward recursive self-improvement (RSI).

Speech transcripts used as long-term memory must preserve both words and stable speaker identities.

A transformer language model performs a bounded amount of computation per token, and recent work by Vishal Sikka, former CEO of Infosys, argues that this bound limits which tasks a model can carry out or verify (arXiv:25…

A robot that observes people interacting with objects should be able to carry out later requests that refer back to those interactions.

Real-world embodied tasks, from everyday activities to professional procedures, require agents to act under physical constraints while tracking evolving object and task states.

Large language models often solve independent problems in the same conversation.

Large language models are strong general-purpose code generators, but executable algorithmic trading remains a demanding specialization target: a model must translate a natural-language strategy specification into correc…

Recent advances have enabled unified omni-modal models in understanding audio, vision, and language.

Delegating complete application development to coding agents requires preserving the intended design rather than simply producing plausible outputs through naive prompting.

Foundation models can serve as clinical agents through tool-use harnesses.

Agent harnesses often improve language models with human-designed workflows, but as models grow more capable, unreliable guidance can increasingly constrain their execution.

Autoregressive (AR) grounding models serialize spatial predictions, introducing sequential latency and imposing a causal order on output tokens.

Precise grounding matters.

Given a target dataset, such as faces with eyeglasses, and a background dataset, such as faces without, contrastive analysis separates salient factors specific to the target from common content shared by both.

Self-attention gives LLMs fine-grained, query-dependent access to context, but dense token interactions incur quadratic prefill cost and a key--value cache growing with context length.

Long-horizon robot manipulation reuses skills across many task compositions, but improving these compositions with additional end-to-end demonstrations is costly.

On-policy self-distillation (OPSD) trains mathematical reasoning models using a privileged teacher that sees a reference solution and supervises student-sampled prefixes.

Latent communication enables multi-agent systems to exchange information directly in internal representation space, reducing the token, computation, and latency overhead of text-based communication.

Vision-language-action (VLA) models generalize broadly across robotic manipulation tasks, but complex environments require balancing task success with unintended contact.

Pre-pretraining (PPT) on synthetic non-natural language data improves token efficiency during language model pre-training (PT).

Reconstructing dynamic driving scenes from recorded sensor data supports closed-loop evaluation of autonomous driving systems by synthesizing observations beyond the original trajectory.

Scientific software presents a demanding test for computer-using agents based on visual language models (VLMs): completing a research workflow requires interpreting specialized interfaces, manipulating scientific objects…

Hour-scale audio-visual question answering is constrained by a context dilemma: dense whole-recording encoding rapidly exhausts context limits, whereas uniform temporal compression severely dilutes fine-grained acoustic…

Reference listening is a common strategy in music production, but current comparison tools often obscure a key human judgment: deciding what should be compared.

Terminal agents act through stochastic model generations, yet the ability to generate a useful action does not ensure its reliable execution.

Ultra-long video temporal grounding requires balancing long-range evidence search with fine-grained event understanding under a limited visual budget, yet existing agentic methods still rely largely on predefined policie…

Existing auto-research benchmarks often entangle multiple sources of improvement, including training frameworks, hyperparameters, compute budgets, and data, making it difficult to attribute why one frontier agent outperf…

An unsuccessful LLM agent rollout contains more information than its final reward: the observations available to the agent, the actions it chose, and the environment's responses.

Modern transformers pair impressive capabilities with substantial memory and compute demands.

We present Game-Guided Skill Discovery (GGSD), a framework that uses self-play in games to discover motor skills that are directly playable by humans.

Video generation models (VGMs) offer strong spatiotemporal priors for embodied observation--action modeling.

Long-term egocentric video enables personalized AI assistants to reason about daily life.

Visual perception is conventionally formulated as a one-shot prediction from a single glance at the image, under the assumption that the image content and the model's parametric knowledge suffice to resolve the query.

Large language models (LLMs) have shown promise for recommendation reranking, but their use introduces an important tradeoff between recommendation quality and serving efficiency.