AI News — Saturday, May 2, 2026

Meta has acquired a robotics startup, signaling a significant move to strengthen its efforts in humanoid AI development and expand its presence in the robotics sector.

Replit CEO Amjad Masad discusses the recent Cursor deal, ongoing disputes with Apple, and his long-term vision for the company, providing insights into the competitive landscape of AI-powered development tools.

This paper introduces a framework for collaborating diverse scientific foundation models, aiming to enhance complex scientific problem-solving through interdisciplinary AI integration.

Researchers propose a new paradigm for visual generation, shifting from simple atomic mapping to advanced agentic world modeling, promising more sophisticated and coherent visual outputs.

Google details how its Tensor Processing Units (TPUs) are engineered to handle the escalating demands of complex AI workloads, highlighting their role in advancing AI infrastructure.

OpenAI shares a reflective piece on the origins and evolution of certain AI behaviors or 'goblins' within their models, offering a glimpse into the challenges of AI development and control.

OpenAI highlights Choco, a company leveraging AI agents to streamline and automate the complex processes of food distribution, demonstrating practical AI application in logistics.

A new research paper introduces ExoActor, a method for generating exocentric video that enables generalizable interactive control of humanoid figures, pushing boundaries in virtual character animation and robotics.

Researchers present RoundPipe, a technique designed to enable efficient AI model training across multiple consumer-grade GPUs, making advanced training more accessible.

Claw-Eval-Live is introduced as a new live agent benchmark designed to evaluate and improve AI agents' performance in dynamic, real-world workflow environments.

This paper unveils Nemotron 3 Nano Omni, an efficient and open multimodal AI model, aiming to provide accessible advanced intelligence across various data types.

A new framework is proposed to guide the confident migration of Large Language Models (LLMs) in production systems when they reach their end-of-life, addressing a critical operational challenge.

Blender announces upcoming changes to its Development Fund and new policies regarding AI integration, signaling its stance on AI's role in open-source 3D content creation.

A developer recounts rebuilding Karpathy's NanoChat using JAX, offering a detailed analysis of XLA's strengths and weaknesses in practical AI development.

A user expresses frustration over their LLM assistant's repetitive use of the phrase "be completely candid," highlighting a common challenge in refining AI conversational patterns for better user experience.