The Deep Dive: NotebookLM's take on Newsletter #8

This week’s episode breaks down the biggest AI news and trends from mid-August 2025. We cover strict new safety rules, the debate around human review, the rise of AI companions, powerful new tools, and surprisingly tiny AI models that can run offline.Timestamps00:00 – 00:55 Introduction: Why AI news feels nonstop00:55 – 03:40 AI safety and governance: Anthropic’s new rules, political content policies, and Google’s stance on AI in search03:40 – 04:47 Human review and the risk of AI training loops04:47 – 06:28 AI companions: benefits, risks, and mental health concerns06:28 – 09:42 AI tools in action: Runway video suite, Google Flights, LinkedIn hiring tools, Meta advertising09:42 – 12:39 Tiny but powerful: Multiverse Computing’s Superfly and ChickBrain, Google’s Gemma 3-170M, Liquid AI’s compact vision models12:39 – 15:11 Incremental vs radical: GPT-5, Gartner’s hype cycle, and Salesforce’s CoActOne system15:11 – 17:20 Claude Sonnet 4’s million-token context and OpenAI’s new modes and integrations17:20 – 19:07 Google Gemini memory updates and Anthropic’s Socratic learning mode19:07 – 19:41 AI + robotics: AI2’s MOMO Act for 3D reasoning19:41 – 21:16 Wrap-up and final reflection: balancing innovation, safety, and human-centered AI Get full access to The AI Vaults at theaivaults.substack.com/subscribe