AI News — Saturday, November 14, 2026
The departure of Nvidia CEO Jensen Huang is reportedly creating significant uncertainty and a funding gap for AI startups, according to venture capitalists and founders.
A new benchmark, SO-Arena, is introduced to evaluate the capabilities and reliability of 'System One' AI models, which are designed for rapid, intuitive responses.
AI agent development platform LangChain has achieved a significant milestone, reaching a $3 billion valuation in its latest funding round.
Reports indicate that export-restricted Nvidia chips are still being utilized to power China's most advanced AI models, raising concerns about trade regulations.
OpenAI has launched ChatGPT Wallet, a new feature designed to integrate financial transactions and services directly within the ChatGPT platform.
Researchers propose PhysLoop, a method that leverages Large Language Models' understanding of physics to enable self-improvement in video generation models.
The winners of the Hacktoberfest Open-Source AI Challenge Week 4, focusing on 'Ghost in the Machine' themes, have been announced, celebrating community contributions to open-source AI.
A new technique called G-STAR is introduced to achieve deeply compressed autoregressive generation by efficiently scaling token capacity.
Open Vision Banana presents a comprehensive and open-source framework for generative visual perception, offering a new recipe for the community.
This research explores the different roles and decision-making processes within long-horizon AI agents, categorizing them as actors, judges, or teachers.
The popular image background removal service remove.bg is shutting down, prompting the community to share free, privacy-focused alternatives that don't require uploading images.
Ropedia Xperience 2 is introduced as a new benchmark and environment for effectively scaling physical AI simulations and complex agent behaviors.
A new paper investigates the fundamental design principles of hyper-connections in neural networks, focusing on simplification, stability, and scalability.
TTB (Test-time Baking) is proposed as an efficient rendering technique for decoder-only view synthesis models, improving performance during inference.
A guide demonstrates how to run the Gemma 4 31B model on AMD Token Lab for free, including setting up a local chat interface and an ADK agent.