All 2,608 AI stories, newest first · page 97 of 109
A new arXiv paper quantifies how floating-point precision issues in large language models lead to chaotic behavior. The research highlights the need for better numerical stability in AI systems.
Researchers introduce SciFi, a new agentic AI framework designed for safe, autonomous execution of scientific tasks. The system combines isolated environments and self-assessment mechanisms to enhance reliability in research applications.
Researchers introduce RiskWebWorld, a realistic benchmark for evaluating GUI agents in high-stakes e-commerce risk management. It features 1,513 tasks from production risk-control pipelines, addressing a gap in current benchmarks.
Researchers introduce ReSS, a hybrid framework that merges symbolic and neural models to improve tabular data prediction. The approach aims to enhance both accuracy and human-understandable reasoning in high-stakes domains like healthcare and finance.
OpenAI has introduced GPT-Rosalind, a specialized AI model designed to enhance drug discovery, genomics analysis, and protein reasoning. This model aims to accelerate scientific research workflows in the life sciences sector.
A new study introduces conformal prediction to quantify uncertainty in large reasoning models, addressing gaps in traditional methods. This approach provides statistically rigorous uncertainty sets, crucial for complex reasoning tasks.
Researchers have developed a method to quantify exploration and exploitation in language model agents without accessing their internal policies. This breakthrough could improve AI decision-making in complex tasks.
Researchers find that the primary bottleneck in scaling multimodal large language models (MLLMs) is knowledge density in training data, not task format. Task-specific supervision like Visual Question Answering (VQA) adds little incremental semantic information beyond image captions.
Researchers introduce GeoAgentBench, a dynamic evaluation framework for LLM-based geographic information systems (GIS). It addresses gaps in static testing by assessing real-time, multimodal spatial analysis capabilities.
Fleeks is a new infrastructure platform designed to remove bottlenecks for AI agents, enabling them to execute, verify, and integrate code seamlessly. The platform aims to bridge the gap between code generation and real-world application.
Researchers introduce CONCORD, a framework for privacy-aware AI collaboration. It enables assistants to work together while only capturing the owner's speech, addressing key privacy concerns.
Researchers introduce a lightweight parallel monitoring system to detect and recover from reasoning degradation in LLM agents. The Cognitive Companion reduces overhead compared to existing solutions. The study shows promise in mitigating issues like looping and drift in multi-step tasks.
Canva's AI 2.0 update brings advanced prompt-based editing to its design suite, aiming to centralize AI-powered content creation. The update includes new tools and capabilities for users to create and adjust designs through natural language prompts.
Anthropic has released Claude Opus 4.7, its most powerful generally available model yet. The update enhances capabilities in software engineering, image analysis, and instruction following. This launch comes amid growing anticipation for the upcoming Mythos Preview.
Reports suggest Trump officials are encouraging banks to test Anthropic's Mythos model, despite the Department of Defense labeling the company a supply-chain risk. This move highlights the ongoing tension between political and military AI stances.
The latest installment in the 'LLM from scratch' series covers gradient accumulation, a technique for training large models on limited hardware. This method enables efficient local training of a 32K-parameter model.
Researchers introduce compute-grounded reasoning (CGR), a paradigm where sub-problems are solved deterministically before language models generate answers. Spatial Atlas implements CGR to tackle complex spatial and machine learning benchmarks.
Researchers propose Self-Distillation Zero (SD-Zero), a method that improves training efficiency by converting binary rewards into dense supervision. This approach outperforms traditional reinforcement learning methods in verifiable settings.
A new paper proposes a two-counter system to evaluate AI memory quality by tracking co-occurrence with success or failure. This could revolutionize how agents manage and prioritize memories over time.
A new study on Llama 3.1 8B Instruct shows that the cognitive core of an agent exhibits attractor-like behavior in activation space. This suggests persistent architecture in large language models. The findings could reshape our understanding of LLM cognition.
OpenAI's chief revenue officer, Denise Dresser, sent a four-page memo to employees outlining the company's strategic direction. The memo emphasizes the need to lock in users and grow its enterprise business to combat competition, particularly from Anthropic.
OpenAI has updated its Agents SDK with native sandbox execution and a model-native harness, enabling developers to build secure, long-running agents. This advancement simplifies the integration of AI agents across various files and tools.
A new AI support chatbot allows users to upload markdown docs and get cited answers. The entire backend is a single JavaScript file, requiring no infrastructure. It uses Next.js, OnCell, and OpenRouter.
Researchers propose a method to improve factuality in long-form LLM outputs by calibrating reasoning confidence. This approach reduces overconfident incorrect claims without post-hoc revision.