Researchers have developed a method that enables AI models to achieve gold-medal-level reasoning capabilities in Olympiad-style problems through simple and unified scaling techniques.
OpenAI has introduced a new personal finance experience within ChatGPT, allowing users to connect their bank accounts for personalized financial insights and management.
Databricks is integrating OpenAI's advanced GPT-5.5 model to power sophisticated enterprise agent workflows, signaling a major step in AI adoption for business automation.
A new method, Causal Forcing++, enables scalable, real-time interactive video generation using few-step autoregressive diffusion distillation, significantly improving efficiency.
New research introduces Self-Distilled Agentic Reinforcement Learning, a technique that allows AI agents to learn and improve autonomously through self-distillation.
MemLens offers a new benchmark for evaluating the long-term memory capabilities of large vision-language models across various multimodal tasks.
Silicon Valley's energy infrastructure is under strain, requiring new providers as the escalating power demands of AI data centers significantly increase electricity prices.
Researchers present SANA-WM, an efficient model capable of minute-scale world modeling using a novel hybrid linear diffusion transformer architecture.
MemEye introduces a visual-centric evaluation framework designed to rigorously test and benchmark the memory capabilities of multimodal AI agents.
The Darwin Family method proposes MRI-Trust-Weighted Evolutionary Merging to scale language model reasoning without requiring additional training.
A comprehensive survey explores the complexities of collaboration, failure attribution, and self-evolution within multi-agent systems powered by large language models.
The STALE research investigates how LLM agents can determine the validity of their memories, addressing a critical challenge in long-term agent autonomy.
WildClawBench is introduced as a new benchmark specifically designed for evaluating AI agents in real-world, long-horizon tasks, pushing the boundaries of practical agent assessment.
An article advises on selecting appropriate AI models, emphasizing that larger models are not universally superior and practical considerations should guide choices.
A project demonstrates building a multimodal Gemini agent, named "Sweets Vault," that integrates with physical hardware for real-world interactions.