Oh io dovrei pure essere in patria, quasi quasi

I have been going through the same path and outcome for an early career grant, so I feel you 100%.

The network is bipartite (only heterosexuals people were interviewed) and the enrolled partners were not asked to nominate their partners, but just to enumerate them, so I am afraid the observations stop there.

Community structure could be looked at instead, as we ask where people live.

Thanks! 🙏

The data comes from a sexual health questionnaire + some contact tracing, where they tried to enroll as many contacts of the indexes as possible. These people were also given the same sex questionnaire.

The questionnaire asks about sexual behaviour, number of partners over a year, risky behaviour..

Congratulazioni 🎉

Thank you 🙏

Manlio, this went out literally now 🤣

"Are LLMs about to emulate human behavior in cooperative dilemmas on networks? These results highlight a crucial gap: LLMs struggle to emulate the nuanced, adaptive social strategies humans deploy in fixed networks"

arxiv.org/abs/2411.10294

www.sciencedirect.com/science/arti...

arxiv.org/abs/2307.04986

There is also some code to it

github.com/bear96/GABM-...

There's already a number of papers on arxiv showing that it "kind of" works. I personally think that there's a number of big problems to address

1) reproducibility
2) the need of "small" language models if you have one specific population (culture/religion...)
3) scalability
4) what do you "fit"?

Tangential to this: when using agent-based models, e.g. to simulate epidemics, would you replace a set of simple, mechanistic rules that you coded/control yourself, with a set of LLM agents that exhibit much more complex behaviour? (let's forget about the reproducibility issue with LLM for now)

Following up on this because it is quite fascinating

Wooooooow ben ritrovata!

I wonder why it is centered around a 1000, and if it has moved in the last few months.

Sono più nell'infectious disease modelling che nell'epi applicata, ma consiglierei questo starter pack qui

bsky.app/profile/scau...

Quando ero meno applied, in un paper avevamo usato l'idea che la dinamica di un SIR/SIS a livello di popolazione "somiglia" a un birth death process per fare inferenza, quindi quando è uscito l'articolo abbiamo aperto una discussione sui limiti e le proprietà di tale approssimazione...