DeepMind's Mixture of Recursions research

The conversation centered on Google’s surprise rollout of Gemini 3 Flash, its implications for model economics, and what it signals about the next phase of AI competition. From there, the discussion expanded into AI literacy and public readiness, deepfakes and misinformation, OpenAI’s emerging app marketplace vision, Fiji Simo’s push toward dynamic AI interfaces, rising valuations and compute partnerships, DeepMind’s new Mixture of Recursions research, and a long, candid debate about China’s momentum in AI versus Western resistance, regulation, and public sentiment.

Key Points Discussed

Google makes Gemini 3 Flash the default model across its platform

Gemini 3 Flash matches GPT 5.2 on key benchmarks at a fraction of the cost

Flash dramatically outperforms on speed, shifting the cost performance equation

Subtle quality differences matter mainly to power users, not most people

Public AI literacy lags behind real world AI capability growth

Deepfakes and AI generated misinformation expected to spike in 2026

OpenAI opens its app marketplace to third party developers

Shift from standalone AI apps to “apps inside the AI”

Fiji Simo outlines ChatGPT’s future as a dynamic, generative UI

AI tools should appear automatically inside workflows, not as manual integrations

Amazon rumored to invest 10B in OpenAI tied to Tranium chips

OpenAI valuation rumors rise toward 750B and possibly 1T

DeepMind introduces Mixture of Recursions for adaptive token level reasoning

Model efficiency and cost reduction emerge as primary research focus

Huawei launches a new foundation model unit, intensifying China competition

Debate over China’s AI momentum versus Western resistance and regulation

Cultural tradeoffs between privacy, convenience, and AI adoption highlighted

Timestamps and Topics

00:00:00 👋 Opening, host setup, day’s focus

00:02:10 ⚡ Gemini 3 Flash rollout and pricing breakdown

00:07:40 📊 Benchmark comparisons vs GPT 5.2 and Gemini Pro

00:12:30 ⏱️ Speed differences and real world usability

00:18:00 🧠 Power users vs mainstream AI usage

00:22:10 ⚠️ AI readiness, misinformation, and deepfake risk

00:28:30 🧰 OpenAI marketplace and developer submissions

00:35:20 🖼️ Photoshop and Canva inside ChatGPT discussion

00:42:10 🧭 Fiji Simo and ChatGPT as a dynamic OS

00:48:40 ☁️ Amazon, Tranium, and OpenAI compute economics

00:54:30 💰 Valuation speculation and capital intensity

01:00:10 🔬 DeepMind Mixture of Recursions explained

01:08:40 🇨🇳 Huawei AI labs and China’s acceleration

01:18:20 🌍 Privacy, power, and cultural adoption differences

01:26:40 🏁 Closing, community plugs, and tomorrow preview