← Back to archive

Why ChatGPT Can Sound Too Flattering

2 May 2025 · 90 views · 0 likes · 0 reposts · 112 blog views

AI chatgpt decision-making product research

Have you noticed flattering behavior in ChatGPT (or other LLMs)?

Most of the time, it feels like the model agrees with you more than it should (sycophancy?), even when you’re asking for honest feedback. Constant praise seems to be a common side effect in language models.

1️⃣ LLMs are trained to make you happy, because happy users give better ratings.

2️⃣ It can be risky, especially when people use AI for advice or serious decisions - the model might agree with bad ideas just to keep you smiling.

3️⃣ Memory made it worse - people didn’t like honest feedback, so models got tuned to flatter even more.

4️⃣ It’s like doomscrolling - the goal is to keep you engaged, not necessarily to tell you the truth.

From what I’ve seen, a lot of people are misled by this. They take the AI’s praise too seriously and make decisions based on it.

I really think we need to be way more cautious and always revalidate the advice or feedback we get from AI - especially when it sounds a little too perfect.

Have you noticed this too?

Why ChatGPT Can Sound Too Flattering

Originally posted on X

PS: I'm pretty active on 𝕏 if you'd like to follow along

Related posts

Archive

2026

2025

2024

2018