Science – Page 152

Beyond Words: How AI Builds Its Own Languages

29.01.2026 by ebaster

The study demonstrates that large language models-specifically GPT, Qwen, and Pixtral-exhibit distinct distributions of natural and task-oriented language variants when applied to the Flags dataset, as revealed through a UMAP projection, suggesting inherent biases in how these models process and generate language for differing communicative goals.

New research reveals that vision-language models are capable of developing task-specific communication systems, diverging from human language in surprising ways.

Decoding Dynamics: When Neural Networks Meet Equation Discovery

29.01.2026 by ebaster

A predictive model, initially trained on only one second of cart-pole dynamics and two up-shift datasets from a bio-model, successfully extrapolated predictions five times beyond its training duration-and even generalized to entirely unseen down-shift events-demonstrating an ability to discern and replicate underlying system dynamics despite limited exposure and challenging extrapolation tasks.

New research shows how continuous-time neural networks can enhance our ability to model and understand complex systems from limited data.

Beyond Black Boxes: Uncovering Relationships in Tabular Data

29.01.2026 by ebaster

The architecture learns contextual relationships through adaptive gated nodes, ultimately yielding predictions via a hybrid functional committee bolstered by a residual linear bypass-a design prioritizing graceful degradation rather than resisting the inevitable entropy of complex systems.

A new neural network architecture prioritizes interpretability and efficiency by explicitly modeling the mathematical connections between features in structured data.

Robots with Memory: Building Smarter Missions

29.01.2026 by ebaster

The methodology unfolds as a cyclical process, each stage-depicted as a discrete node-yielding tangible outputs contained within defined boundaries, suggesting an iterative refinement inherent to the system’s growth.

New research details a robust methodology for equipping robots with knowledge graphs, enabling more intelligent and adaptable mission execution.

From Scribble to Screen: Guiding AI Animation with Free-Form Sketches

29.01.2026 by ebaster

SketchDynamics facilitates the iterative refinement of animation sequences through proactive clarification and user-driven specification, evidenced by a storyboard depicting traffic signals-where initial sketches of red, yellow, and green lights are expanded with details regarding vehicle movement timing and velocity-ultimately yielding a finalized, dynamic video clip.

Researchers have developed a new interface that lets users communicate animation ideas to artificial intelligence through intuitive, hand-drawn sketches and iterative refinement.

Beyond Understanding: Making AI Explanations Truly Useful

29.01.2026 by ebaster

User accountability frameworks categorize information and associated actions, implicitly forecasting potential points of failure within the system's operational logic.

A new framework assesses how well AI explanations empower users to take meaningful action, shifting the focus from interpretability to practical impact.

Robots That Remember: Scaling Collaboration with Shared Experience

29.01.2026 by ebaster

$The system operates through a layered planning process, initially guided by a large language model →, then extended through specialized modules [latex] \Rightarrow [/latex] to refine and expand upon the initial strategic framework.$

A new framework enables teams of robots to learn from past tasks, dramatically improving efficiency and reducing the need for constant re-planning.

Smart Buildings Get a Brain Boost

29.01.2026 by ebaster

A new agentic AI framework promises to unlock significant energy savings and automation in building operations.

Robots Learn by Asking: The Rise of AI-Powered Manipulation

29.01.2026 by ebaster

New research shows that artificial intelligence agents can learn to control robots through trial and error, without needing pre-programmed demonstrations.

From Concept to Conclusion: Automating the Scientific Story

29.01.2026 by ebaster

The system cultivates research directions not by building solutions, but by growing a knowledge graph from existing work, allowing underspecified intent to be grounded in concrete patterns retrieved and composed from this curated ecosystem of reusable method units.

A new framework streamlines research by converting initial ideas into complete scientific narratives, accelerating discovery and knowledge synthesis.