AI News — Saturday, May 9, 2026

10
Cloudflare says AI made 1,100 jobs obsolete, even as revenue hit a record high

Cloudflare announced that AI automation led to the obsolescence of 1,100 jobs, highlighting the significant impact of AI on employment despite the company's record revenue growth.

TechCrunchindustry
9
Skill1: Unified Evolution of Skill-Augmented Agents via Reinforcement Learning

Researchers propose Skill1, a framework for the unified evolution of skill-augmented agents using reinforcement learning, enabling more complex and adaptable AI behaviors.

Hugging Faceresearch
9
MiniCPM-o 4.5: Towards Real-Time Full-Duplex Omni-Modal Interaction

The MiniCPM-o 4.5 model aims to achieve real-time, full-duplex omni-modal interaction, representing a significant step towards more natural human-AI communication.

Hugging Faceresearch
9
Testing ads in ChatGPT

OpenAI is reportedly testing advertisements within ChatGPT, indicating a potential shift in its monetization strategy and user experience.

OpenAI Blogproduct
8
Beyond Semantic Similarity: Rethinking Retrieval for Agentic Search via Direct Corpus Interaction

A new paper explores moving beyond traditional semantic similarity in retrieval systems for agentic search, advocating for direct interaction with the corpus to improve results.

Hugging Faceresearch
8
Continuous Latent Diffusion Language Model

A novel approach introduces the Continuous Latent Diffusion Language Model, pushing the boundaries of generative AI for language tasks.

Hugging Faceresearch
8
The Local Model That Doesn't Sleep: Gemma 4 + MTP as a Marathon Engine

This article explores the potential of Gemma 4 combined with MTP to create a continuously running local AI model, suggesting a shift towards more viable on-device intelligence.

Dev.toopen-source
7
MiA-Signature: Approximating Global Activation for Long-Context Understanding

Researchers introduce MiA-Signature, a method for approximating global activation to enhance long-context understanding in AI models, crucial for processing extensive information.

Hugging Faceresearch
7
RaguTeam at SemEval-2026 Task 8: Meno and Friends in a Judge-Orchestrated LLM Ensemble for Faithful Multi-Turn Response Generation

RaguTeam's participation in SemEval-2026 Task 8 showcases a judge-orchestrated LLM ensemble designed to generate more faithful and coherent multi-turn responses.

Hugging Faceresearch
7
When to Trust Imagination: Adaptive Action Execution for World Action Models

This paper investigates the critical question of when AI agents should trust their 'imagination' and adapt action execution within world action models, improving agent autonomy.

Hugging Faceresearch
7
MARBLE: Multi-Aspect Reward Balance for Diffusion RL

MARBLE presents a new approach for Multi-Aspect Reward Balance in Diffusion Reinforcement Learning, aiming to optimize learning across various reward signals.

Hugging Faceresearch
7
Why does AI lie? Hallucinations explained simply

An article provides a simple explanation for why AI models 'lie' or hallucinate, demystifying a common challenge in AI development and usage.

Dev.toindustry
7
SkillOS: Learning Skill Curation for Self-Evolving Agents

SkillOS proposes a system for learning skill curation, enabling AI agents to self-evolve and adapt their capabilities over time.

Hugging Faceresearch
6
I Built My Mom an AI Recipe Helper for Mother's Day

A developer shares a personal project of creating an AI-powered recipe helper for Mother's Day, demonstrating practical and accessible AI applications.

Dev.toproduct
6
Using Claude Code with Docker Model Runner

A practical guide demonstrates how to integrate and use Claude's code generation capabilities with a Docker Model Runner, streamlining development workflows.

Dev.toproduct