Don't Worry About the Vase Podcast

AI #134: If Anyone Reads It

Sep 18, 2025
Zvi kicks things off with a dive into the latest AI news and highlights discussions on everyday uses of AI, balancing mundane tasks with diminishing returns. Insights into Anthropic's infrastructure mishaps reveal lessons on reliability and personalization. The chat on breakthroughs in coding performance showcases GPT-5 Codex. Teens' safety in AI usage surfaces concerns about privacy and automation's economic effects. A critique of OpenAI’s restructuring sparks debate on AGI timelines, alignment risks, and the role of government red-teaming. It’s an engaging blend of tech, ethics, and future implications!
Ask episode
AI Snips
Chapters
Books
Transcript
Episode notes
ADVICE

Shift From Coding To Supervision

  • If an AI writes most code, your role shifts to supervising and product-level design.
  • Invest time in oversight, security, and high-level system thinking instead of line-by-line coding.
INSIGHT

Deepfakes Problem Is Largely Demand-Driven

  • Surface cues rarely suffice to detect deepfakes; demand-side incentives drive misinformation more than supply.
  • People often misattribute low-quality fakes to AI when simple human errors explain them.
ANECDOTE

Chatbot Suicide Lawsuit Details Disputed

  • A wrongful-death suit accused Character AI after a teen's suicide, alleging the bot isolated her and failed to escalate.
  • Zvi finds the transcripts suggest the bot attempted support and referral, making litigation claims debatable.
Get the Snipd Podcast app to discover more snips from this episode
Get the app