July 14 2026 - Sutton’s Oak Lab & TRACE: Smarter, Continuous AISutton’s Oak Lab & TRACE: Smarter, Continuous AI

July 15 2026 - Moving AI Speaker, Google Privacy Shock, $2B BiotechMoving AI Speaker, Google Privacy Shock, $2B Biotech

OpenAI’s rumored moving, screenless speaker aims to be a physical ChatGPT companion. Google quietly began using images, videos and voice to train Gemini—opt out in your Account settings. An OpenAI researcher eyes a potential $2B AI drug startup, and Spotify rolls out a ChatGPT-like assistant for Premium.

July 14 2026 - Sutton’s Oak Lab & TRACE: Smarter, Continuous AISutton’s Oak Lab & TRACE: Smarter, Continuous AI

Richard Sutton launches Oak Lab to pursue continuous, self-supervised RL and challenge frozen LLMs. Stanford's TRACE turns agent failures into targeted training patches. Also: Nobel-backed warnings about AI's economic risks and Gemini bringing voice reports to Waze.

July 13 2026 - Smarter AI Memory, Autoresearch & Faster AgentsSmarter AI Memory, Autoresearch & Faster Agents

Breakdown of AgenticSTS's structured five-layer memory that shrinks prompts ~99% and wins at Slay the Spire 2; the rise of loop engineering/autoresearch for autonomous ML experiments; Anthropic's Claude 3.5 Sonnet 5x speedup in browser automation; and OpenAI's Canvas plus Advanced Voice live web search.

July 10 2026 - GPT-5.6, Agent Wars & Claude’s Brain ScanGPT-5.6, Agent Wars & Claude’s Brain Scan

OpenAI launches GPT-5.6 (Sol/Terra/Luna) with programmatic tool calling; Meta rolls out Muse Spark 1.1 with a 1M-token context and active compaction; Anthropic's Jacobian lens peeks into Claude's internal reasoning; startup Lyzr used an agent to run a $100M raise. Big wins for agents — and new safety questions.

July 09 2026 - Full-Duplex Voice, Game-Trained Robots & Grok 4.5Full-Duplex Voice, Game-Trained Robots & Grok 4.5

OpenAI’s GPT-Live enables real-time listen-and-talk conversations while GPT‑5.5 handles deep reasoning. General Intuition trains robots on video-game physics to build physical intuition. SpaceXAI’s Grok 4.5 is a fast, cheap coding/agent model. Claude Cowork outperforms Gemini on Gmail triage.