1193 stories curated by AInformed · page 10 of 50
Researchers introduced Just Keep Prompting (JKP), a multi-turn evaluation framework that measures how well Vision-Language Models (VLMs) maintain epistemic stability when users repeatedly challenge, question, or contradict their answers across up to 10 follow-up turns.
A new ArXiv paper proves that large language models (LLMs) have an inherent reliability ceiling that no amount of scaling can overcome. The study decomposes output uncertainty into a resolvable component (closable with more context) and a subjective component (inherent to task ambiguity), and shows that autoregressive generation further degrades this ceiling.
Researchers have enhanced neuro-symbolic AI by adding probability to logical reasoning, enabling AI to make educated guesses about unknown information while preserving its logical structure. This could lead to robots that learn and reason like humans, but with more transparency and structure.
Researchers created a way to measure how well AI tools work together. It helps AI assistants use the right tools at the right time, making them faster and more reliable.
A new arXiv position paper from researchers argues for a shift in AI autoformalization from single statements to complete theories—including axioms, definitions, and lemmas—to create unified, machine-verifiable formal knowledge bases. This could transform how complex knowledge in mathematics, law, and science is structured and validated.
Researchers from ArXiv cs.CL have developed a system that automatically evolves prompt guidelines tailored to specific tasks, addressing the common problem of underspecified user queries. This could make large language models like ChatGPT and Claude more accurate and user-friendly without requiring prompt engineering expertise.
Researchers introduced MAPS (Multi-Agent Perspective Spaces), a framework that enables multiple AI agents to maintain individualized beliefs, emotions, and cognitive styles during dialogue, avoiding the semantic uniformity of current systems. It uses domain-weighted profiles, GRU-based memory, and token-level attention for interpretable, diverse interactions.
A new ArXiv study shows that LLM-based AI agents lose information when communicating via plain text, suggesting they possess an internal 'world model' that exceeds textual expressibility. This finding has implications for multi-agent system design and alignment.
Researchers introduced SPINE (Scalable Physical Integration with ageNtic Expertise), an agentic AI framework that automates the debugging and deployment of bimanual robots, reducing the need for expert-driven calibration and making advanced robotics more accessible.
A new arXiv paper proposes 'networked intelligence' — teams of specialized AI systems collaborating like human scientists — to tackle complex scientific challenges that single models cannot solve alone.
A new study from ArXiv cs.AI introduces a mathematical framework for underwriting and pricing insurance tailored to autonomous AI systems. The model accounts for risks like decision-making, tool use, and environmental interactions, filling a gap left by traditional insurance policies.
Researchers have developed a new system called Oracle Agent Memory to help AI assistants remember important details over long conversations. This could make AI tools more reliable for tasks that span multiple sessions, like complex projects or ongoing customer support.
A new survey on arXiv frames self-improving AI agents as adaptive systems that convert experience into capability gains with minimal human input. The research defines modern agents as a configuration coupling a foundation model with an operational scaffold of prompts, memory, tools, and control logic.
Researchers introduce interventional grounding audits, a black-box method that substitutes predicates in LLM chain-of-thought reasoning to test whether conclusions genuinely depend on stated premises, evaluated on the ProntoQA benchmark.
Researchers introduced Safe-Psych, a sequential benchmark that evaluates how well large language models handle evolving diagnostic uncertainty in clinical psychiatry. Unlike existing medical benchmarks that assume complete information upfront, Safe-Psych tests whether models request clarification or abstain when data is insufficient.
Researchers introduced DROPJ, a human-centered method that trains AI agents safely by combining world models with human feedback and justifications. This approach could make AI systems more reliable in safety-critical domains like healthcare and autonomous driving.
Researchers introduced OriginBlame, a record- and token-level data provenance system that lets data contributors precisely remove their specific content from AI training datasets without over-deleting unrelated data.
Researchers propose a meta-learning framework for RLHF and DPO that transfers preference data across languages, enabling effective alignment of large language models in low-resource languages with minimal data.
FixItFlow is an AI system that automatically generates troubleshooting guides from historical cloud incident data using large language models, aiming to replace manual, labor-intensive documentation and improve incident response consistency.
A new arXiv study reveals that the European Patent Office's 2026 guidelines create a paradox: human-written patent claims, required to be clear and concise under Article 84, can exhibit the same low-perplexity patterns as AI-generated text, making detection nearly impossible on consumer GPUs.
Researchers propose a modular Context-Augmented Prompting framework that combines small language models (SLMs) with graph neural networks (GNNs) to overcome structural blindness in molecular property prediction. The method provides predictive hints and explanatory subgraphs at inference time, improving accuracy for drug discovery and materials science.
Researchers introduce Interactive Multi-Feature Fusion, a method that combines multiple semantic feature spaces to reconstruct thoughts from non-invasive brain recordings with greater accuracy than prior single-dimension approaches.
A new study introduces MAGE (Memory-Augmented Goal-directed Prompt Evolution), a framework that reveals a previously unknown phenomenon in how different components of AI prompt optimization interact, exposing critical trade-offs between stability and performance.
A new graph-based AI framework from ArXiv cs.CL detects and analyzes disinformation narratives on Telegram by clustering related claims and modeling their diffusion across channels. The tool targets coordinated misinformation campaigns, particularly in the Russia-Ukraine conflict.