EP 601 : Dario Amodei Fires Back on AI Safety & OpenAI Solves 22‑Year Math Problem

Ready for the hottest AI headlines of the day? 🎙️ 🔴 **Dario Amodei steps onto X** – Anthropic’s CEO finally replies to the backlash over his safety warnings. He argues that fear‑mongering won’t win trust; only real breakthroughs—like curing cancer—can restore public confidence. Hear Amodei’s take on the “false choice” between lock‑downs and open models, and why he believes Anthropic’s measured approach could protect smaller labs while slowing the biggest players.🧮 **ChatGPT cracks a 22‑year‑old math mystery** – Beijing neurosurgery resident Shanmu Jin used GPT‑5.6 to prove Crouzeix’s Conjecture, a matrix problem unsolved since 2004. We dive into his unconventional setup: a sandboxed model, swarms of sub‑agents, and a 16‑hour autonomous session. What does a doctor‑turned‑mathematician’s success mean for the future of research, peer review, and the role of human mathematicians?💾 **OpenAI rolls out “Computer History”** – A brand‑new opt‑in feature that logs your clicks and keystrokes, giving ChatGPT and Codex a short‑term memory of recent work. We explore the privacy trade‑offs, potential productivity gains, and how this could change the way developers and power users interact with AI assistants. **Why you should listen** – Whether you’re a founder, developer, researcher, or just AI‑curious, this episode gives you the context you need to navigate the fast‑moving landscape. We’ll help you separate hype from real impact, point out the tools that can boost your productivity today, and leave you with clear takeaways you can act on.👍 **Join the conversation** – Hit play, share with your network, and subscribe for weekly deep dives into AI breakthroughs, controversies, and the tools shaping tomorrow’s tech. Your feedback fuels the show, so drop us a comment or tweet your thoughts on Amodei’s safety stance or the ChatGPT proof!**Tools mentioned:** Better Claw, Tenorshare, Astorie, GLM‑5.3, Pika Audio, Computer History, Gemini Flash 3.7.