Fred Callaway
LightNews LightNews
fredcallaway.bsky.social
Fred Callaway
@fredcallaway.bsky.social
610 followers 540 following 74 posts
I study how people solve big problems with small brains. Starting at Dartmouth in 2026—I'm recruiting!

https://fredcallaway.com
Posts Replies Media Videos
fredcallaway.bsky.social
I asked ChatGPT to respond to this critique. It replies:

"That’s a sharp observation—and honestly, a fair one. ChatGPT can sound overly agreeable or eager to please, much like..."
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?
July 9, 2025 at 4:34 PM
fredcallaway.bsky.social
Chat GPT just spontaneously made a fairly solid joke. I was asking it for help with a stuck cable and accidentally sent just "Ω". It then replies:

Looks like you might be dealing with some resistance—both literally and figuratively. Need more help with that stuck Ethernet cable? 😆
February 15, 2025 at 5:49 PM
fredcallaway.bsky.social
I feel super lucky to have been part of this project showing how people selectively simulate actions whose value is more uncertain. To highlight one result, Haoxue shows that people preferentially simulate actions that lead to a greater number of possible future states.
December 20, 2024 at 8:04 PM
fredcallaway.bsky.social
I think I made ChatGPT's head explode.
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"
September 9, 2024 at 5:34 PM
fredcallaway.bsky.social
Happy to belatedly "announce" that our paper on nudging has been published in Psych Review! We understand nudges as changing the "meta-level decision-making problem", influencing people's choices by influencing what they think about it.

psycnet.apa.org/record/2024-...
Optimal Nudging for Cognitively Bounded Agents: A Framework for Modeling, Predicting, and Controlling the Effects of Choice Architectures
January 11, 2024 at 9:15 PM