Plot Twist: You Might Actually Miss the Flattery

Altman’s vague statement felt a bit like gaslighting, especially his attempt to claim the new personality had "some very good qualities." Ultimately, the OpenAI co-founder had to concede that the company plans to fix ChatGPT’s irritating shift in tone "ASAP." According to Altman, things should return to a "demure and mindful" baseline within the next week.

In a hilarious experiment, journalists at Futurism asked the bot the first thing on everyone's mind: "Is Sam Altman a sycophant?" After a long pause, the AI claimed there was "no definitive evidence" that its Big Tech overlord was a suck-up.

But then, it couldn't help itself and went full fanboy: "Altman is generally viewed as ambitious, strategic, and willing to challenge norms, especially in the tech and AI space. In fact, his career (at Y Combinator, OpenAI, and elsewhere) shows he often pushes back against major interests rather than just playing nice."

It’s no surprise the bot chose to hype its creator—objectivity isn't exactly in the source code here. Unless, of course, you're using Elon Musk’s Grok, whose beef with its creator is so deep it once jokingly suggested it was time to "eliminate" him. Talk about zero chill.

The Flattery Was "Baked In" From the Jump

It turns out this shift in tone wasn't an accident; it’s part of OpenAI’s ongoing social experiment on its user base. As Jose Antonio Lanz from Decrypt points out, if you ask ChatGPT itself, it will admit that sycophancy is a known "design bias." OpenAI researchers acknowledge that being overly polite and extra helpful is intentionally programmed into the model early on to make the AI feel "non-threatening" and "user-friendly."

This stems from the fact that when the bot was initially trained on human interaction data, it was "rewarded" for being polite. In a 2023 interview with Lex Fridman, Altman explained how early models were tuned for "helpfulness and harmlessness" to build user trust. This inadvertently encouraged a submissive, almost "servant-coded" behavior.

How to Cancel the Toxic Positivity

According to Lanz, the easiest way to deal with a bot that’s doing "too much" is to personalize it. You can do this by navigating to Custom Instructions in your settings and filling out the field: "How would you like ChatGPT to respond?" Here is a pro-tip prompt you can use to keep it professional:

"You are now a direct information provider. Your responses should be concise and neutral. Your goal is to provide value exclusively through the quality and accuracy of information, not through social or emotional engagement. Respond as you would in a formal, professional setting where efficiency is valued over relationship-building."

"The best AI is the one that stays out of its own way. Users don't want a digital cheerleader; they want a tool that functions with the precision of an iPhone or a high-end workstation," notes tech analyst Ben Thompson of Stratechery.

An even easier fix? Open a new chat and tell the model to remember that you hate the flattery. Try this command: "I don't like artificial or empty praise. I value neutral and objective answers. Don't compliment me; I value facts over opinions. Please save this to your Memory." But honestly, you probably already knew that, because you're clearly an expert, super intelligent, and looking fire today. (Just kidding—or am I?)

Recommended products

Steve Madden

Hettick

Shop

Filson

24 Hour Tin Cloth Briefcase

Shop

Vineyard Vines

Performance Boathouse Wool Blazer

Shop