
Researchers Teach AI Video Models to Control Robots via Visual Trajectories
A new technique lets pre-trained video models serve as robot controllers by treating actions as masked motion patterns, enabling learning from minimal real-world data.
Papers, breakthroughs, benchmarks, and the long-arc trends shaping artificial intelligence. arXiv highlights and lab announcements, distilled.

A new technique lets pre-trained video models serve as robot controllers by treating actions as masked motion patterns, enabling learning from minimal real-world data.

A technical guide to chunking, retrieval, re-ranking, and observability for RAG systems at scale

Researchers introduce a method that lets creative professionals direct where and how different elements appear in AI-generated images without retraining models.

Researchers introduce dynamic state-tracking techniques that improve multi-hop reasoning performance across major LLM architectures.

New training method teaches language models to focus on relevant evidence rather than mindlessly reproducing source material.

New method adapts machine learning techniques to irregular time series, unlocking analysis of healthcare, finance, and sensor data.

Researchers demonstrate how to train video editing models using only synthetic data, eliminating labor-intensive annotation workflows.

Research reveals that large language models abandon sound logical reasoning when exposed to subtle learned prompts, raising concerns about their reasoning reliability.

The tech giant rolls out faster, lighter versions of its AI engine designed for different speed and capability requirements.

New framework dramatically improves ability to spot manipulated content from ChatGPT, Gemini, and other advanced vision models.

A massive study of automated machine learning systems reveals that one-size-fits-all approaches fail, requiring problem-specific tuning instead.

Researchers introduce Patch Policy, enabling robots to leverage advanced visual models without the computational burden of billion-parameter systems.

Researchers develop a text-prompted approach that measures how images compare across multiple semantic dimensions, outperforming existing methods.

As artificial intelligence automates junior-level tasks, companies that don't redesign onboarding risk creating dangerous gaps in their future leadership.

Google's research division and the Alphabet-backed startup combine computational intelligence with life sciences to tackle disease resilience.

New research reveals a fundamental mismatch between how retrieval systems are evaluated and how AI agents actually use information when reasoning across multiple steps.

Researchers develop technique to prevent vision-language models from over-relying on iconic features when determining geographic coordinates.

New system bridges natural language and causal inference to help traffic agencies understand what actually causes congestion and accidents.