1100 stories tagged Research · page 17 of 46
Researchers created SEATauBench, the first framework to test AI agents in Southeast Asian languages. It evaluates how well AI tools work in Mandarin, Vietnamese, Thai, Indonesian, and Filipino across progressively localized settings that change the language of user interaction, tool specs, and task domains.
A new research paper introduces Odyssey, a categorical framework for building AI models that preserve local truths through composable 'foundries.' This could make AI more reliable in specialized contexts like law or medicine.
OpenAI introduced GeneBench-Pro, a new benchmark for evaluating AI performance in genomics and biology. It uses real-world datasets to test how well AI models handle complex scientific problems.
Researchers studied how AI models make mistakes when planning in networked environments. Their findings could improve AI tools that manage interconnected systems like supply chains or social networks.
Researchers developed a new approach called ATOD to improve AI agents for complex tasks. It combines imitation learning with reinforcement learning to help agents learn faster and perform better.
Researchers introduced SD-GPS, a solver-driven framework that addresses key bottlenecks in geometry problem solving: autoformalization and theorem prediction. By treating the symbolic solver as an execution oracle, SD-GPS improves accuracy and flexibility, with potential applications in math education.
Researchers have introduced Yuvion, a large language model specifically built to handle adversarial attacks. This could make AI systems safer by preventing harmful or misleading outputs, especially in scenarios involving planning, tool use, and multi-step reasoning.
A new study shows that AI models using quick, instinctive responses outperform slower, more deliberate thinking for recognizing emotions. This challenges the assumption that more reasoning always leads to better accuracy.
Researchers found that the personality traits of AI agents in teams can impact how well they work together. This could help design better AI collaboration tools for the future.
Researchers compare two approaches to world models for AI agents—agent-based and parameterized—showing that parameterized models reduce hallucination propagation by using measurable errors, leading to more reliable planning in step-by-step tasks.
Researchers suggest creating a collaborative network for AI models to reduce training costs and deployment challenges. This could make advanced AI tools more accessible to smaller organizations and individuals.
Researchers developed a new method to improve AI's ability to plan complex tasks. This could make AI assistants more reliable for long-term decision-making.
Researchers created a new benchmark called DiscoBench to evaluate how well AI search tools ask clarifying questions when users' requests are unclear, vague, or even factually incorrect. This helps AI assistants avoid cascading errors in multi-step searches.
Researchers have developed a new way for AI to simulate future outcomes, making it better at long-term planning. This could help AI agents make decisions more like humans do, by imagining different scenarios before acting.
Researchers introduced Tree of Evidence (ToE), an AI system that builds argument trees to fact-check claims. It dynamically gathers and evaluates evidence from multiple sources to combat misinformation, including AI-generated content under Generative Engine Optimization (GEO) poisoning.
Researchers created a benchmark to test if AI planners can follow hidden social norms, not just explicit goals. This helps AI act more appropriately in everyday environments, like knowing not to interrupt people.
Researchers created DysLexLens, an AI framework to study how dyslexic learners use AI tools. It analyzes online forum discussions to better understand their needs and experiences.
A new position paper argues that the term 'machine unlearning' is overused in LLM research. It should be reserved for precisely removing the influence of a specific dataset from a model. Many current applications of the term are misleading and dilute its meaning. (2026-06-29)
Nous Research announced Hermes MoA (Mixture of Agents) virtual models, which scored 8% higher than Opus 4.8 and 11% higher than GPT 5.5 on key AI benchmarks. This breakthrough could lead to better performance in everyday tasks like writing and problem-solving.
A new study from AllenAI reveals that hybrid AI models predict certain types of tokens better than others, with key implications for training more accurate language models.
Larger AI models consistently outperform smaller ones in reasoning tasks. Researchers developed a new tool to study why this happens, revealing key differences in problem-solving approaches.
A new research paper reveals that the classical intuition that verifying a solution is easier than producing one is being inverted for today's coding agents. As foundation models get stronger, generating candidate solutions has become easier, while reliable verification—capturing underspecified human intent—has become the harder problem.
Researchers argue that after a benchmark's accuracy saturates, the focus should shift to six other key dimensions of AI performance: construct validity (shortcuts), out-of-distribution generalizability, efficiency, reliability, model vs. scaffold importance, and human–AI collaboration uplift.
A new paper highlights flaws in how we test AI models that handle text, images, and other inputs together. Current methods miss key aspects like understanding physical reality or combining different types of information. The authors propose better evaluation frameworks to address these gaps.