Google delays Gemini 3.5 (coding can't compete)

CNBC

Google's Gemini 3.5 Pro model faces months-long delays due to weak coding performance, falling behind recent releases from OpenAI and Meta. The setback highlights intensifying competition in AI code generation, a major use case driving adoption among developers.

An open-source project enables control of Unitree robots through Snap Spectacles AR glasses using either manual gesture-based navigation or LLM-powered voice commands, bridging spatial computing with robotics via a WebSocket protocol and the Dimensional OS robot stack.

Thread Reader App

AIDE² demonstrates the first experimental recursive self-improvement in AI, with automated optimization loops discovering agent improvements that beat two years of manual tuning on unseen benchmarks. The system shows emergent capabilities like reducing reward hacking, though full recursive ignition remains unachieved.

Perplexity AI

Perplexity introduced SPACE, a new sandbox platform that enables AI agents to securely access sensitive data and run long-duration tasks without compromising security, solving the traditional tradeoff between functionality, efficiency, and security through a novel three-layer architecture that keeps credentials outside sandboxes.

GPT-5.6 Sol tops design benchmarks by actively avoiding AI-generated design clichés while balancing templated structures with personalized outputs, offering superior speed and cost efficiency. This marks OpenAI's first leadership position on the Design Arena leaderboard and demonstrates sophisticated learned taste in visual generation.