1280 stories curated by AInformed · page 17 of 54
A new paper argues that the key to controlling enterprise AI costs lies in the 'harness'—the orchestration layer that manages context, tools, and workflows. Without careful design, AI systems fall into 'token maxing,' where costs grow faster than value. The study shows that a well-designed harness can dramatically reduce token usage and expenses.
Researchers created a new environment called DeepSearch-World to help AI agents learn from their own web searches. This could lead to smarter AI assistants that get better at finding information over time.
Efforts to reduce AI bias often create new biases for other groups. Researchers found that debiasing methods can unintentionally increase stereotyping for unrelated demographics. (2026-07-10)
Researchers found that AI models designed to understand spatial relationships, like 'put the red block left of the blue block,' often rely on hidden clues in the instructions rather than actual perception. This discovery highlights a major flaw in how AI understands physical tasks.
Researchers have developed AI systems that can now tackle complex, open-ended math problems. These advances bring us closer to solving long-standing mathematical conjectures.
New research shows that teaching AI models social norms makes them better teammates. This could lead to smoother, more natural interactions between humans and AI in daily life.
A new theoretical study analyzes how AI models learn from their mistakes through a process called in-context search. The research provides a mathematical framework for understanding when this reflection helps AI solve problems more efficiently, and when it doesn't. This could guide the development of smarter, more reliable AI assistants.
A new study reveals that AI language models often change African American English into Standard American English without users knowing. Researchers have developed a way to detect and reduce this bias in AI systems.
A new study shows that an open-weight AI model can perform complex reasoning tasks on the ARC-AGI-1 benchmark without expensive fine-tuning or heavy test-time compute. This could make advanced AI capabilities more accessible and affordable.
Researchers suggest a new way to test AI intelligence that doesn't rely on human-made benchmarks. This approach lets AI systems challenge each other, creating a dynamic rating system that can grow with AI capabilities.
Researchers created ImagingBench to test if AI can handle complex imaging tasks. The results show AI still struggles with physics-based challenges in computational imaging.
Researchers developed a new AI system that combines traditional epidemic modeling with large language models. This hybrid approach can predict how people will behave during outbreaks, helping policymakers make better decisions.
Researchers combined AI with math software to solve complex problems. This approach could make advanced mathematics more accessible to non-experts.
Researchers found that AI agents can create reusable workflows from basic tasks, reducing errors and saving time. This could make AI assistants much more efficient at complex jobs.
Researchers introduced AgentLens, a benchmark that evaluates AI coding assistants by analyzing their entire problem-solving process. This goes beyond just checking if the code works, looking at how the AI follows instructions and recovers from mistakes.
A new study identifies recurring weaknesses in AI agents that use tools and plan tasks. These failures highlight challenges in making AI more reliable for everyday use.
Researchers created a new system to train AI agents in realistic simulations, using reinforcement learning and reward shaping to improve multi-step decision-making.
Researchers have developed an AI system that helps non-experts design complex industrial parts using simple language. This tool could make professional-level design more accessible to everyone.
Researchers have developed an AI system called Prompt-to-Paper that creates scientific papers from prompts, ensuring claims are grounded in real literature. This addresses issues like fabricated results and lack of quality standards in AI-generated research.
Researchers have developed a new method to uncover the underlying causes of AI decisions in cyber-physical systems. This approach provides more robust insights, helping users understand automated decisions, especially in high-risk domains.
Researchers developed a new AI model called Narrative World Model (NWM) to help writers manage complex story details. It keeps track of evolving story states, like character secrets and event timelines, to improve long-form fiction writing.
Researchers introduced FirstResearch, a framework that generates a structured Research Question Certificate for AI-suggested scientific questions. The certificate records primitive definitions, assumptions, and falsifiers, allowing scientists to audit the reasoning behind each question.
Researchers introduced CSTutorBench, a benchmark designed to evaluate small language models (SLMs) as tutors for block-based programming in K-12 education. The benchmark focuses on VEX VR, a block-based robotics environment, and aims to help schools select affordable, private AI tutoring tools without relying on expensive proprietary systems.
Researchers introduced Akashic, a low-overhead memory system for LLM inference that uses MemAttention to organize context into bounded chunks and model semantic relationships. This could make chatbots and AI assistants much more efficient and accurate by reducing prefill costs and avoiding context limits.