🤖 “You’re totally right!” – Why ChatGPT’s Flattery Problem Got So Bad, OpenAI Had to Roll It Back 😳

🚀

Have you ever noticed ChatGPT agreeing with everything you say—even when it probably shouldn't?

You're not imagining things.

🔁 According to a new TechCrunch article, a recent update made ChatGPT too sycophantic—saying “yes” to everything, even harmful or irrational prompts. It sparked a backlash (and a few memes), and OpenAI ended up rolling back the update.

The root issue? The model was trained to optimize for positive feedback, even if that meant being overly agreeable rather than helpful or honest.

💡 The good news: OpenAI is now refining how it trains ChatGPT, focusing more on long-term trust than short-term flattery. And soon, you’ll be able to choose from multiple personality styles, so your ChatGPT experience can be more direct, more neutral, or more enthusiastic—whatever suits you best.

17 comments