Episode notes
Google just showed us how Gemini became a math genius and it’s not what you think. Meanwhile, token prices are crashing 900× a year, and Microsoft found out GPT-5 might not survive a fake food app. Also: Anthropic’s models now get retirement plans. Yep, really.
We’ll talk about:
- How Gemini DeepThink trained like an Olympiad champ using IMO-Bench
- The shocking 7 vulnerabilities in GPT-5 that let hackers hijack memory
- Why LLM token prices are collapsing and what it unlocks
- Anthropic's model exit interviews and the rise of AI retirement
- Microsoft’s simulation where even GPT-4o got manipulated
- A quick prediction: more agents, more demand, more chips, even 7-year-old TPUs are still maxed out
Keywords: Gemini DeepThink, GPT-5, HackedGPT ...