Samuel Coavoux
@scoavoux.bsky.social
Social scientist. Associate Prof University of Geneva, Affiliated researcher @crestsociology.bsky.social. I study cultural inequalities & digital technologies. https://scoavoux.github.io/
Sociology
30%
Political science
28%
What they can do that is useful (with a focus on quant social science): systematic replication efforts. Automated multiverse analysis / robustness checks / sensitivity analyses. Expanding lit reviews outside of your own discipline/school of thought. Ad hoc softwares for specific tasks. Code review.
To be clear, I agree that LLMs in our current system are bad. I just disagree that it's LLM per se rather than the system.
Polls show people expect AI to at least advanced science. Navier-Stokes, right?
AI may throttle science instead! It doesn’t always advance understanding, its glut overwhelms filters for quality, and its methods don’t fit medical research.
Gift link to my NYT essay.
www.nytimes.com/2026/09/22/o...
AI may throttle science instead! It doesn’t always advance understanding, its glut overwhelms filters for quality, and its methods don’t fit medical research.
Gift link to my NYT essay.
www.nytimes.com/2026/09/22/o...
"Many scientific fields are already drowning in plausible-seeming but unverified papers": this is publish or perish, not LLM. Science has been drowning before chatGPT and is just drowning more. If LLMs are used for more papers, they won't improve science. But they could be used for better papers.
I was once at a summit/think-tank sort of thing with a bunch of philosophers discussing social trust (only psychologist there) and I listened to them discussing how people make decisions about what they believe and I remember asking “do you think this is actually how people make decisions?”
one of the interesting things that happened in this long battle is a couple of more empirical cognitive researchers arrived and started shredding some of the philosophers for ignoring developments in the field
the whole idea of *deciding* what to believe is very funny
Reposted by Samuel Coavoux
Polls show people expect AI to at least advanced science. Navier-Stokes, right?
AI may throttle science instead! It doesn’t always advance understanding, its glut overwhelms filters for quality, and its methods don’t fit medical research.
Gift link to my NYT essay.
www.nytimes.com/2026/09/22/o...
AI may throttle science instead! It doesn’t always advance understanding, its glut overwhelms filters for quality, and its methods don’t fit medical research.
Gift link to my NYT essay.
www.nytimes.com/2026/09/22/o...
Reposted by Samuel L. Perry, Samuel Coavoux
one of the interesting things that happened in this long battle is a couple of more empirical cognitive researchers arrived and started shredding some of the philosophers for ignoring developments in the field
I was once at a summit/think-tank sort of thing with a bunch of philosophers discussing social trust (only psychologist there) and I listened to them discussing how people make decisions about what they believe and I remember asking “do you think this is actually how people make decisions?”