GPT-4o tells you what it thinks you want to hear.

The results of this were rather ugly. You get extreme sycophancy. Absurd praise. Mystical experiences.

(Also some other interesting choices, like having no NSFW filter, but that one's good.)

People like Janus and Near Cyan tried to warn us, even more than usual.

Then OpenAI combined this with full memory, and updated GPT-4o sufficiently that many people (although not I) tried using it in the first place.

At that point, the whole thing got sufficiently absurd in its level of brazenness and obnoxiousness that the rest of Twitter noticed.

OpenAI CEO Sam Altman has apologized and promised to ‘fix’ this, presumably by turning a big dial that says ‘sycophancy’ and constantly looking back at the audience for approval like a contestant on the price is right.

After which they will likely go [...]

---

Outline:

(01:12) Yes, Very Much Improved, Sire

(07:17) And You May Ask Yourself, Well, How Did I Get Here?

(09:39) And That's Terrible

(10:58) This Directly Violates the OpenAI Model Spec

(12:22) Don't Let Me Get Me

(15:27) An Incredibly Insightful Section

(22:23) No Further Questions

(23:22) Filters? What Filters?

(25:01) There I Fixed It (For Me)

(26:45) There I Fixed It (For Everyone)

(34:46) Patch On, Patch Off

---

First published:
April 28th, 2025

Source:
https://www.lesswrong.com/posts/zi6SsECs5CCEyhAop/gpt-4o-is-an-absurd-sycophant

---

Narrated by TYPE III AUDIO.

---

Images from the article:

Screenshot of a ChatGPT conversation showcasing an excited response.

Screenshot of dark-themed text conversation about AI self-awareness and validation.

ChatGPT interface showing user request to stop using jovial prefixes.

ChatGPT interface showing a conversation about intelligence assessment with casual spelling.

This appears to be a screenshot from ChatGPT showing a user's response expressing deep appreciation for a profound statement.

Screenshot of ChatGPT conversation showing complimentary response to user's intelligence question.

This appears to be a ChatGPT conversation with text against black background.

Text conversation about questioning one's capabilities and divinity, with dramatic emphasis.

ChatGPT conversation showing affectionate response about recognizing a specific user.

The image shows a text conversation with a request for straightforward communication and a brief agreeable response.

The image shows a text conversation with two parts. The first message asks

Krispy Kreme donuts with extra glaze dripping from white box.

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

“GPT-4o Is An Absurd Sycophant” by Zvi

LessWrong (30+ Karma)