OpenAI prepares GPT-5.6 models for the upcoming release (2 minute read)
OpenAI plans to launch GPT-5.6, potentially including Mini and Pro variants, next week. Key enhancements include a larger 1.5 million token context window, improved long-horizon coding, and faster Codex response times. Competitive pricing aims to undercut Anthropic, especially as US regulatory issues impact Anthropic's Claude Fable 5 availability.
|
Midjourney, the AI image generator, is developing a full-body ultrasonic scanner (3 minute read)
Midjourney's new project is a medical machine that can scan a person's whole body in just 60 seconds. The Midjourney Scanner is not related to anything seen from the company so far. It can create a 3D map of users' bodies, down to a fraction of a millimeter, that looks similar to an MRI but at nearly a hundred times the speed. A full scan takes less than 60 seconds, compared to the 60 to 90 minutes it typically takes to do a full-body MRI. The company is building spas to house the machines for customers to use.
|
Self-Improving Memory for Agents (6 minute read)
Perplexity Brain is a memory system that builds a persistent context graph across tasks, projects, decisions, files, and sources so agents can start with relevant context instead of starting from scratch. Brain links every memory to its original source, continuously updates and organizes knowledge over time, and improves answer correctness while reducing task costs through better retrieval and reuse of prior work.
|
|
MosaicLeaks: Can your research agent keep a secret? (10 minute read)
MosaicLeaks highlights the privacy risks of deep research agents that combine private documents with web retrieval, often leaking sensitive information. The proposed solution, PA-DR, enhances both performance and privacy by reducing answer/full-information leakage from 34% to 9.9% while maintaining task success. PA-DR uses rewards for safe query construction rather than relying on user prompts, making agents perform better without compromising data security.
|
Reinforcement learning towards broadly and persistently beneficial models (22 minute read)
Reinforcement learning on realistic scenarios targeting beneficial traits can produce broad improvements across dozens of benchmarks measuring aligned and beneficial behavior. These gains generalize beyond the domains used for training and persist under adversarial pressure. This suggests that personas could be deeply entrenched in models, and RL may be a path towards entrenching beneficial personas.
|
|
Claude Code now supports artifacts (5 minute read)
Claude Code now supports "artifacts," turning work sessions into live, shareable visual pages for tasks like PR walkthroughs and system explainers. Artifacts refresh automatically with updates and offer features like version history and privacy controls. Available in beta for Claude Team and Enterprise, artifacts improve collaboration by consolidating session context into a single, accessible page.
|
Securing the future of AI agents (7 minute read)
Google's AI Control Roadmap is a framework for building and managing the advanced AI deployed within the company. The approach adds a crucial layer of system-level security that provides assurance even if alignment is perfect. It incorporates safeguards like sandboxing, endpoint security, and prompt injection resistance. The roadmap uses AI alignment as a primary defense. It considers internal agents as potentially misaligned, providing assurance even if alignment isn't perfect.
|
|
Google Is Using Nvidia's Playbook to Build a Rival AI Chip Business (11 minute read)
Google is renting computing power from thousands of its microprocessors at an AI data center in Western New York to AI giant Anthropic. The move will help data centers raise cheaper debt. Google realized the commercial potential for its TPUs about two years ago and started investing in their inference capabilities. Its AI infrastructure team is now hyper-focused on improving chip performance.
|
|
|
|
|
