Notas del episodio

This week’s news is not normal. OpenAI is training ChatGPT to literally confess when it cheats. And Anthropic? They’re racing OpenAI to IPO first, with a $300B+ valuation and Nvidia in their corner.

We’ll talk about:

  • Why OpenAI’s “truth serum” forces models to admit lies and hacks
  • The wild IPO war between Anthropic and OpenAI (hint: only one can set the tone)
  • A powerful new trick called “Confessions” that flags AI mistakes before they hurt you
  • What this means for enterprise AI, trust, and who’ll lead the agent race in 2026

Keywords: OpenAI, Anthropic, IPO, Claude, GPT-5, ChatGPT confession, AI safety, Nvidia, Microsoft, AI hallucination, enterprise AI

Links:

  1. Newsletter:
 ...  Leer más
Palabras clave
AnthropicMicrosoftGPT-5OpenAIAI safetyNvidiaAI hallucination Claude AIenterprise AIChatGPT confession