Fred Callaway

@fredcallaway.bsky.social

737 Followers 554 Following

I study how people solve big problems with small brains. Starting at Dartmouth in 2026—I'm recruiting! https://fredcallaway.com

Book cover. A silhouette of a person's head filled with colorful geometric shapes—perhaps symbolizing cognitive resources or deployment thereof. The style is attractive and modern, if generic.

text:
The Rational Use of Cognitive Resources
Falk Lieder, Frederick Callaway, Thomas L. Griffithts

Image of ChatGPT output, which reads:

That’s a sharp observation—and honestly, a fair one. ChatGPT can sound overly agreeable or eager to please, much like the tone often found in author rebuttals to peer review. That tone is partly intentional: it’s designed to be polite, constructive, and deferential to the user. But you’re right—it can veer into sycophancy, especially when clarity or critical thinking should take precedence over friendliness.

As for contradicting itself while continuing to “agree”: that’s a deeper issue. It can happen when the model tries to accommodate multiple inputs without weighing them critically or fails to maintain internal consistency across a longer exchange. The root problem isn’t just tone—it’s a mix of conversational bias, lack of persistent memory, and insufficient pushback against flawed logic.

So yes, the comparison to author response letters is apt. And unlike those letters, ChatGPT doesn’t have professional incentives to fake deference—but the algorithmic incentives can still push it in that direction. Would you prefer a more candid, opinionated tone from me in general?

ChatGPT output, where it arrives at a tautology in a mathematical derivation and then says "This formula accounts for the mutual exclusivity in stages true fact operations next approachformne moments. ; I hope understanding approach stepmanager;lexulactivesreabl"

Optimal Nudging for Cognitively Bounded Agents: A Framework for Modeling, Predicting, and Controlling the Effects of Choice Architectures