AI News — Sunday, May 3, 2026

The Academy of Motion Picture Arts and Sciences has announced that AI-generated actors and scripts will no longer be eligible for Oscar awards, setting a new precedent for creative industries.

This paper introduces a novel method for policy distillation that co-evolves policies, potentially improving efficiency and performance in reinforcement learning.

This paper discusses the concept of 'agent-native research artifacts,' suggesting a future where AI agents play a primary role in generating scientific papers and research outputs.

This paper demonstrates an end-to-end autonomous system capable of scientific discovery on a real optical platform, showcasing AI's potential in experimental science.

Intern-Atlas proposes a new research infrastructure designed to map the methodological evolution of AI, aiming to help scientists navigate and contribute more effectively to the field.

This research explores the use of verifier-based reinforcement learning to enhance image editing capabilities, allowing for more precise and controlled modifications.

This paper presents a scalable value pretraining approach for token-level length modeling, which could improve the control and predictability of output length in large language models.

This research proposes Diffusion Templates, a unified plugin framework that enhances the controllability of diffusion models for various generative tasks.

FAMA introduces a failure-aware meta-agentic framework designed to improve the robustness and reliability of open-source LLMs in interactive tool-use scenarios.

The NHS is reportedly taking a stance against open-source solutions, raising concerns about the future of open technology adoption in public healthcare.

Google's Gemini 3.1 introduces native text-to-speech capabilities, enhancing summary reading with more powerful and accessible audio features.

This article highlights a free alternative named 'Goose' that reportedly offers similar code generation capabilities to Claude Code, which costs up to $200 monthly, potentially disrupting the market for AI coding assistants.

This paper introduces a new Representation Fréchet Loss designed to improve the quality and fidelity of visual generation models.

This research investigates the use of synthetic computers at scale to simulate long-horizon productivity, offering a new approach to evaluate and develop AI agents.

This article provides a practical guide on building a persistent AI assistant using Google Cloud Engine, Hermes Agent, and Telegram, demonstrating real-world application of cloud AI services.