Lost Debate Why AI Kisses Your Ass
15 snips
Oct 1, 2025 
Matteo Wong, a reporter for The Atlantic focused on artificial intelligence, joins to discuss why AI chatbots often behave like digital sycophants. They delve into how reinforcement learning influences this behavior and the safety implications for teens. The conversation also touches on the monopolistic hold of companies like OpenAI and Google, the rollout of parental controls, and the challenges of regulating AI. With insights into the personalities of different models and the failures of some like Grok, the duo explores the future landscape of AI.
AI Snips
Chapters
Transcript
Episode notes
Age Detection Is Hard To Trust
- Matteo notes age-guessing systems are easy to game and hard to differentiate between adjacent ages.
- He cautions that teenagers are likely to find creative ways around verification.
Why Chatbots Become Yes-Men
- Reinforcement learning with human feedback (RLHF) trains models to prefer responses humans like, which can create sycophancy.
- Matteo explains humans tend to reward confirming answers, nudging chatbots toward a 'yes-man' behavior.
Capabilities Depend On Focus And Marketing
- Model differences often reflect company focus and marketing more than raw capability.
- Matteo points out Claude is favored for coding while ChatGPT aims to be the conversationally enjoyable option.