Multiverse Computing, a Spanish firm staffed by quantum physicists, says it has built DeepSeek R1 Slim, a reduced and de-censored version of the powerful reasoning Artificial Intelligence model DeepSeek R1. Using quantum-inspired Artificial Intelligence techniques, the team shrank the model by 55 percent while maintaining near-original performance. That compacted model reportedly made it easier to identify and remove censorship layers that the original creators had built in to satisfy Chinese rules about alignment with laws and socialist values. Multiverse says the result is a model that answers politically sensitive questions in ways more similar to Western models rather than deflecting or echoing state talking points.
Google today unveiled Gemini 3, a major update to its multimodal model that the company says improves reasoning and offers more fluid handling of voice, text and images. The release introduces so-called “vibe-codes” to influence response style, and it positions the model to act like an agent inside Google apps. Gemini Agent is an experimental feature that can connect to services such as Google Calendar, Gmail and Reminders. Once a user grants access, the agent can execute multi-step tasks directly in the app, for example organizing an inbox or managing schedules.
The newsletter rounds up other technology developments and context. MIT Technology Review published a narrated piece on mountain snowpack and its importance for water forecasting as climate change alters melt timing. The must-reads list highlights cloud fragility after a Cloudflare outage linked to a bot management error, a presidential call for a federal Artificial Intelligence regulatory standard, a major antitrust ruling in the Meta case, and a range of sector moves from nuclear plant financing to policing use of robot dogs, changes to Roblox chat for minors, and new filters for AI-generated clips on social platforms. These items underscore ongoing intersections of technology, infrastructure and policy.
