EP 618 : Meta Muse Spark 1.3 & Google Gemini 3.8 Flash: New AI Models, Safety Loops & Productivity Tools

<p>September is already buzzing with AI breakthroughs, and this episode dives into the headline‑making releases from Meta and Google, plus the latest safety debate at OpenAI. We start with Meta’s Muse Spark 1.3 (Max) – a 62‑score model that Zuckerberg calls “frontier performance almost too cheap to meter.” Hear why the upcoming “Watermelon” model could shake the industry and how Meta is opening its weights for the community. Then we shift to Google’s Gemini 3.8 Flash, a flash‑forward from the 2026 slump that keeps Gemini 3.7 pricing while climbing to a 59 on the Intelligence Index. DeepMind’s own Koray Kavukcuoglu weighs in on the frontier race, giving you insider context on Google’s comeback strategy.Next, we unpack OpenAI’s controversial “Astra” loops. The new recurrent‑depth technique squeezes extra intelligence out of the model without inflating size, but it also hides raw reasoning behind pure math outputs. Chief Scientist Jakub Pachocki warns of a “race into unmonitorability,” and we discuss what that means for AI safety monitoring across the ecosystem.We also cover the rest of today’s AI news: Anthropic’s background computer‑use features for Claude, the U.S. Commerce Secretary’s vote of confidence in Anthropic, the NYT copyright brief supporting OpenAI, NYC’s AI ban for grades K‑8, and Spain’s Multiverse Computing debuting Quasar 438B, the top European model on the AI Index.Finally, we spotlight six productivity‑boosting tools that can help you experiment with these models right now: Gemini 3.8 Flash, Muse Spark 1.3, H3 Max Turbo, CinLink, Miligram, and Katto.Tune in, share with your tech‑savvy friends, and hit subscribe so you never miss a beat in the fast‑moving AI arena.Tools mentioned: Gemini 3.8 Flash, Muse Spark 1.3, H3 Max Turbo, CinLink, Miligram, Katto.</p><p><br></p>