The Newcomer Podcast Inside Anthropic’s Plan to Fix AI’s BIGGEST Problem
58 snips
Nov 17, 2025 
Mike Krieger, Chief Product Officer at Anthropic, discusses the need for truth-seeking AI to combat misinformation, emphasizing the importance of quality over quantity in outputs. He explains how structured prompts improve AI interactions. Mati Staniszewski, CEO of ElevenLabs, shares insights into the future of voice AI, highlighting its role in customer support and integration with real-world systems. They delve into safety measures for licensed voices and the importance of authenticity in voice technology, all while tackling challenges like deepfakes.
AI Snips
Chapters
Books
Transcript
Episode notes
Conversational Models Can Reward Verbosity
- Models optimized for conversational quality can drift toward verbosity and engagement rather than accuracy.
- Anthropic warns that eval leaderboards reward length and not necessarily truth-seeking behavior.
What 'Slop' Looks Like
- 'Slop' refers to low-effort, unedited AI outputs that lack critical human thought.
- Mike Krieger frames anti-slop as producing outputs that feel thoughtfully curated, not just first drafts.
Force Iteration And Better Prompts
- Iterate with AI and supply well-constructed prompts to reduce low-quality outputs.
- Build UIs that let models ask clarifying questions and let humans narrow wide option spaces.
