Decoding the Language Machine
A new review systematically maps the evolving theory behind large language models, offering a lifecycle perspective on their development and behavior.
Science
Robots Learn to Navigate Clutter with a Gentle Touch
![A refinement to the control boundary formulation enables realistic safety reasoning in dense environments, allowing for nuanced interaction with objects-such as slight contact and small displacements-where a conservatively defined boundary [latex] \sigma = 0.01 [/latex] would otherwise induce unnecessary stalling.](https://arxiv.org/html/2601.02686v1/x2.png)
A new framework enables robots to safely interact with objects in crowded spaces by learning how to subtly nudge and avoid collisions.
Beyond Speed: Why Reasoning Needs a Sense of Time
New research suggests that imbuing AI agents with the ability to dynamically manage their reasoning process – much like human cognition – is key to unlocking superior problem-solving abilities.
Beyond Turing: Teaching AI to Sound Like Us
Researchers have developed a new framework to quantify and instill human-like qualities in artificial intelligence, moving beyond simple task completion to genuinely natural conversation.
Seeing is Knowing: AI Turns Images and Text into Formal Proofs
Researchers have developed a new framework that leverages the power of artificial intelligence to translate visual and textual information into rigorously verifiable logical statements.
The AI Interview Gap: When Automation Meets Applicant Experience
A new study examines the disconnect between expectations and reality in AI-driven hiring, revealing how current systems often undermine candidate agency and satisfaction.
Beyond Personhood: The Legal Status of Superintelligent Machines
As artificial intelligence rapidly advances, the question of legal responsibility and moral standing for truly intelligent systems demands a critical re-evaluation of existing legal frameworks.
When Robots Listen: Securing the Future of AI-Powered Machines
![Backdoor attacks targeting large language model-driven robotics represent a critical vulnerability, wherein subtle manipulations of input prompts can induce unintended and potentially hazardous robot behaviors, exploiting the LLM’s reliance on statistical correlations rather than robust semantic understanding of the physical world and task objectives [latex] \implies [/latex] a deviation from predictable, safe operation.](https://arxiv.org/html/2601.02377v1/diagrams/backdoor.png)
As large language models increasingly take the reins of physical robots, a critical need emerges to understand and mitigate the unique security vulnerabilities this integration creates.
The Mathematics of Will: Modeling Self-Determined Action
A new mathematical framework proposes how agency can arise from physical systems without violating deterministic laws.
Seeing Beyond the Surface: Holotomography’s Rise in Biological Imaging
Holotomography is rapidly evolving from a niche morphometric technique into a powerful, AI-driven platform for label-free, 3D multimodal phenotyping across diverse biological applications.