Thanks to Salima Jaoua and Christoph Mayer for their constructive feedback !

A joint work with @nicolo-pagan.bsky.social Nicolò Pagan, PhD, Giordano De Marzo, Taehee Kim, @hannahmetzler.bsky.social Hannah Metzler (May), @silviagiordano.bsky.social Silvia Giordano, Aniko Hannak and @dgarcia.eu David Garcia.

Well... 🥁 Good news for the ones using generative ABM! both models tested happen to do a good job in that 😉 particularly, Llama 3 , which fooled participants 44% of the time 😈.

Would you recommend, dropping this path and the generative ABM for social media simulations, or do you still see value for trying out in the case of validating GABM?

But also while reading your paper, it felt like, it's quite risky to invest time and ressources to find some operational validity method (also because of restricted data access).

Your paper gave me a small ideas on how to potentially tackle validation for GABM for social media (something in the line of testing some change in the algorithm and see whether it align with changes abserved in empirical data), but I quickly see important challenges to it (data availability).

I am a PhD student (hopefully in my final year), currently interested in digging deeper into the validation aspect of generative ABMs for social media simulations.

Very timely and valuable contribution ! I really enjoyed the argumentation in it, how it put things into perspective by comparing to the traditional ABM and social science literature.