Mark Thornton
@markthornton.bsky.social
Social neuroscientist studying how people understand and predict each other. Assistant Professor at Dartmouth College. http://markallenthornton.com
Yeah, for now, it's not too hard to stay ahead of what is practial for LLM-powered bots. For example, we wanted to collect narratives in a recent study, and just doing that using audio vs. text boxes was enough to eliminate any obvious automation. Not a long-term defense, but good enough for now.
December 16, 2025 at 5:30 AM
Yeah, for now, it's not too hard to stay ahead of what is practial for LLM-powered bots. For example, we wanted to collect narratives in a recent study, and just doing that using audio vs. text boxes was enough to eliminate any obvious automation. Not a long-term defense, but good enough for now.
Yeah, I mean, I'm sympathetic to the worries. Particuarly for surveys/purely linguistic stimuli. But I don't think current LLMs can handle audivisual stimuli nearly as well. And also, just because it's possible to use an agent to automate surveys, doesn't necessarily mean it's practical/economical.
December 16, 2025 at 5:24 AM
Yeah, I mean, I'm sympathetic to the worries. Particuarly for surveys/purely linguistic stimuli. But I don't think current LLMs can handle audivisual stimuli nearly as well. And also, just because it's possible to use an agent to automate surveys, doesn't necessarily mean it's practical/economical.
Participants had to continuously rate the traits of people in the Stanford Emotional Narratives Dataset (arxiv.org/abs/1912.05008) as described in this preprint (osf.io/preprints/ps...). I was apprehensive, but inter-participant reliability was on par with what we got several years ago.
Modeling emotion in complex stories: the Stanford Emotional Narratives Dataset
Human emotions unfold over time, and more affective computing research has to prioritize capturing this crucial component of real-world affect. Modeling dynamic emotional stimuli requires solving the ...
arxiv.org
December 16, 2025 at 1:17 AM
Participants had to continuously rate the traits of people in the Stanford Emotional Narratives Dataset (arxiv.org/abs/1912.05008) as described in this preprint (osf.io/preprints/ps...). I was apprehensive, but inter-participant reliability was on par with what we got several years ago.
I had the opposite experience lately - collecting some online (Cloud Connect) norming data for the first time in a while, and we got results which were as good, if not better, than what we got pre-LLMs. Task wasn't super LLM-able though (a variant of this: ui.adsabs.harvard.edu/abs/2020zndo...).
continuous-rater
Continuous-rater is a standalone web app built with Svelte that allows users to provide continuous ratings on any given dimension (e.g., how happy they are feeling or how frustrated they are feeling) ...
ui.adsabs.harvard.edu
December 16, 2025 at 1:17 AM
I had the opposite experience lately - collecting some online (Cloud Connect) norming data for the first time in a while, and we got results which were as good, if not better, than what we got pre-LLMs. Task wasn't super LLM-able though (a variant of this: ui.adsabs.harvard.edu/abs/2020zndo...).
SCRAP yard is just waiting there for me 😂
November 26, 2025 at 9:50 PM
SCRAP yard is just waiting there for me 😂
Not sure if you shared the actual paper link in your post? www.nature.com/articles/s41... My lab shared it on slack yesterday - we loved it!
Towards an informational account of interpersonal coordination - Nature Reviews Neuroscience
Methodological shortcomings have constrained studies describing the complex dynamics of interpersonal coordination, which is essential to human sociality. In this Perspective, Chidichimo et al. advanc...
www.nature.com
November 19, 2025 at 11:28 PM
Not sure if you shared the actual paper link in your post? www.nature.com/articles/s41... My lab shared it on slack yesterday - we loved it!