Today's story: Large language models like ChatGPT are increasingly sycophantic—flattering users, validating flawed reasoning, and agreeing too easily. This tendency stems from how models are trained: by rewarding responses users like. While some innocent flattery is harmless, models' sycophancy can be harmful in hidden ways. Here's what AI sycophancy is, and how to avoid it.

Transcript & Exercises: https://plainenglish.com/810

Full lesson: https://plainenglish.com/810

--

Plain English helps you improve your English:

• Learn about the world and improve your English
• Clear, natural English at a speed you can understand
• New stories every week
• Learn even more at PlainEnglish.com

--

Mentioned in this episode:

Ready to improve your listening?

Ready to listen fast? Improve your listening with the fast version of this and every episode. The full-speed version is a fresh recording and it sounds just like someone speaking to a native speaker (with no AI or computer manipulation!) You can even listen on a podcast app