Researchers have described a technique called 'promptomorphism': the generation of endless variations of malicious code by AI tools is making traditional defense methods increasingly ineffective.
Scale AI has launched Voice Showdown, a benchmark for evaluating voice AI models based on real human preferences and live speech.
Lab
Quantum Origami: How to Fold a Perfect Particle State Without Losing It Along the Way
Physics & Space • Quantum Physics
Researchers have proposed a way to synthesize complex quantum states with fewer resources, which could bring us closer to reliable quantum computing.
AI: Events
How AI is Changing Robot Development: From Virtual Environments to Real-World Production
Products
NVIDIA has introduced new tools that integrate simulation, machine learning, and embedded computing into a unified workflow for robot creation.
Mistral has released Small 4, a new compact model that's faster, more accurate, and boasts improved performance across multiple languages, including Russian.
Mistral AI has joined the NVIDIA Nemotron coalition, a partnership aimed at advancing open language models and multimodal AI capabilities.
Rakuten integrated OpenAI's AI agent, Codex, into its development process, significantly reducing the time to resolve failures. This article explains how it works in practice.
AI: Events
How ChatGPT Learns Not to Trust Everything: Protecting Agents from Hidden Commands
Security
OpenAI has shared how it protects ChatGPT agents from hidden instructions within data, explaining why this is crucial as AI begins to act independently.
AI: Events
Agents with an Embedded Computer: OpenAI's Responses API Update
Technical context • Development
OpenAI has updated its Responses API, equipping it with tools that enable autonomous agents to access files, utilize a command line, and operate within an isolated computing environment.