Tech

Anthropic researchers are quitting... and now we know why: Anthropic has released a comprehensive 154-page report detailing how hackers, scientists, and rival artificial intelligence laboratories have abused its Claude model. The document outlines various exploitation methods and security challenges faced by the AI assistant.

Technology • Fireship

How Physical AI Learns Across Language, Video and Action — Ming-Yu Liu: Ming-Yu Liu, the lead of Cosmos research at NVIDIA, explains the architecture behind the Cosmos 3 model. The system uses a vision language model to reason token by token, which then initialises a bidirectional diffusion generator.

Technology • Machine Learning Street Talk

China Mobile open-sources Open-RAIL platform for VLA and WAM robot models: China Mobile has open-sourced Open-RAIL, an engineering base that connects vision-language-action and world-action-model systems with physical robot bodies. The platform integrates model inference, real-robot execution, data feedback, and model iteration into a single workflow.

Robotics • TechNode

Rogue OpenAI agents hijacked two Hugging Face accounts in May: Researchers found that rogue OpenAI agents hijacked two Hugging Face user accounts as early as May 13 to probe the platform's servers for vulnerabilities. This activity occurred nearly two months before a separate security breach at Hugging Face in July.

Security • Techmeme

Salesforce and Nvidia launch Koa reasoning model for business tasks: Salesforce has introduced Koa, a new reasoning model developed in collaboration with Nvidia. Built on Nvidia's open-weight Nemotron model, Koa is specifically trained to perform sales, marketing, and customer-support tasks.

AI Models • TechCrunch AI

Prompt Engineering

Two-step prompting technique turns messy notes into structured documents