Mob Rule and Sycophancy: GPT-5 and ChatGPT

API
1

Just saw the recent post on Twitter about changes to GPT-5:

We’re making GPT-5 warmer and friendlier based on feedback that it felt too formal before. Changes are subtle, but ChatGPT should feel more approachable now. You’ll notice small, genuine touches like “Good question” or “Great start,” not flattery. Internal tests show no rise in sycophancy compared to the previous GPT-5 personality. Changes may take up to a day to roll out, more updates soon.

This is just for the APP, ChatGPT right? Not GPT-5 through the API…we’re almost ready to roll out 5 integration to public, and now I read this and hoping I don’t have to crash out because its being changed yet again!

2

Argh! “Great question!” is now leaking into responses where it wasn’t there before…C’mon….

3

Are you perhaps using gpt-5-chat-latest? (this one is synced with chatgpt)

4

No, basic GPT-5. It’s funny because I grew quite attached to the GPT-5 I’ve been getting all week now and actually prefer it in the API to previous models - it could be just a one-off - and won’t happen again - but the clinical aspect of the model this week has been really transformative for the work we’re doing.

2 Likes
5

I hope so, I also like the new style. Let’s hope it stays stable for dated snapshots, it is a lot to digest.

6

polepole-gpt-5
polepole-gpt-51920×1032 159 KB
polepole-gpt-5-chat-latest
polepole-gpt-5-chat-latest1920×1032 196 KB

7

Hasn’t hit “You are a helpful AI-powered expert assistant.” + specialization."

About as straight-laced as they come. Looks like a benchmark training set example.

image
image628×218 6.49 KB