cocoons_dk retweetledi
New preprint: 𝗧𝗵𝗲 𝗗𝘆𝗻𝗮𝗺𝗶𝗰𝘀 𝗼𝗳 𝘀𝗼𝗰𝗶𝗮𝗹 𝗰𝗼𝗻𝘃𝗲𝗻𝘁𝗶𝗼𝗻𝘀 𝗶𝗻 𝗟𝗟𝗠 𝗽𝗼𝗽𝘂𝗹𝗮𝘁𝗶𝗼𝗻𝘀: 𝗦𝗽𝗼𝗻𝘁𝗮𝗻𝗲𝗼𝘂𝘀 𝗲𝗺𝗲𝗿𝗴𝗲𝗻𝗰𝗲, 𝗰𝗼𝗹𝗹𝗲𝗰𝘁𝗶𝘃𝗲 𝗯𝗶𝗮𝘀𝗲𝘀 𝗮𝗻𝗱 𝘁𝗶𝗽𝗽𝗶𝗻𝗴 𝗽𝗼𝗶𝗻𝘁𝘀
arxiv.org/pdf/2410.08948… 1/N
English
cocoons_dk
34 posts
cocoons_dk retweetledi
Seemingly unbiased LLMs exhibit "collective biases" when interacting with other agents. We ought to look at AI biases *within application contexts*, not simply by probing models in isolation.
Great collab. with @RELflintashery and @a_baronca
@nerdsitu @ITUkbh @Carlsbergfondet
Andrea Baronchelli@a_baronca
We explore how Large Language Models (LLMs) autonomously develop social conventions through interaction, without external programming or centralized coordination. 2/7
English
cocoons_dk retweetledi
@IC2S2 Day 1: fun and insightful poster session focusing on plant-based diets as a climate change mitigation strategy (paper: ojs.aaai.org/index.php/ICWS…) and post environmental disaster group dynamics on Reddit, in collaboration with @lajello 🌱
#ic2s2 @nerdsitu @cocoons_dk
Philadelphia, PA 🇺🇸 English
cocoons_dk retweetledi
LLMs can leverage dimensions of social pragmatics to generate arguments that sound persuasive to humans, and that mimic human persuasion strategies.
@AndersGiovanni presenting @ic2s2.
Paper from #icwsm24 proceedings: arxiv.org/abs/2312.15523 #ic2s2
English
cocoons_dk retweetledi
Testing LLMs in Game Theory experiments reveals their moral biases as social agents -- useful to inform alignment. Llama2 playing Prisoner's Dilemma is more forgiving and non-retaliatory than humans, but only when the opponent defects rarely. New preprint: arxiv.org/abs/2406.13605
English
cocoons_dk retweetledi
🚨 Our paper "The Persuasive Power of Large Language Models" has been accepted to @icwsm 🚨
📖 Check out the pre-print (arxiv.org/abs/2312.15523) or this small post for more details (andersgiovanni.com/blog/post-5-ic…).
See you in Buffalo 🇺🇸✌🏼
#icwsm @nerdsitu @cocoons_dk @Carlsbergfondet @ITUkbh
Anders Giovanni Møller@AndersGiovanni
🤖 Can LLM-based agents interact with each other and generate human-like persuasive arguments? We spice up the agents with social personalities, influencing their persuasion strategies, and ask humans to rate the arguments.
English
cocoons_dk retweetledi
Last week @AndersGiovanni, @lajello, and I had the pleasure of visiting @CentaiInstitute and @ISI_Fondazione to showcase our latest work on the @cocoons_dk project 🌎 Grateful for the valuable feedback and stimulating discussions! Thanks to all for the warm hospitality 🇮🇹
Torino, Piemonte 🇮🇹 English
cocoons_dk retweetledi
Had a wonderful week in Turin with @arianna_pera1 and @lajello, visiting @CentaiInstitute & @ISI_Fondazione. 🇮🇹🍝 We had the chance to present our work on the COCOONS project and got some really thoughtful feedback. 🙌🏼
@nerdsitu @Carlsbergfondet #COCOONS
English
cocoons_dk retweetledi
📣 Doing computational research on how we form opinions on climate change? How does climate awareness spread? When does communication bring change?
Have some early results to present, or published work to discuss?
Join our NO CAPS workshop at @icwsm!
👉 no-caps.github.io/2024/
English
cocoons_dk retweetledi
New preprint out!
The causal role of the Reddit collective action on the GameStop short squeeze. Great work by @ADTony_, with @lajello and @lau_retti
arxiv.org/abs/2401.14999
English
cocoons_dk retweetledi
Ever wondered how storytelling fuels collective action on climate change?🌱
Check out the new pre-print with @lajello: we studied vegan narratives on YouTube and the emergence of collective signals in related reactions arxiv.org/abs/2401.09210
English
cocoons_dk retweetledi
Can artificial agents interact with each other to reproduce human-like persuasive dialogue? And do the arguments they generate sound persuasive to humans?
We used Llama2 to test different persuasion strategies, and asked humans to rate them.
New on ArXiv: arxiv.org/abs/2312.15523
English
cocoons_dk retweetledi
Shifting from YouTube to TikTok, content creators amplify emotional and action language when talking about climate change 📢: tailored communication is the key to steering the climate conversation ⛈️🔥
Check the new pre-print with @lajello: arxiv.org/abs/2312.04974
English
cocoons_dk retweetledi
Video creators can steer the attention of crowds on critical issues like climate change. Understanding how this happens is crucial to refine communication strategies on climate action. New findings from a YouTube vs. TikTok comparison (thanks to the new TikTok Research API)
Arianna Pera@arianna_pera1
Shifting from YouTube to TikTok, content creators amplify emotional and action language when talking about climate change 📢: tailored communication is the key to steering the climate conversation ⛈️🔥 Check the new pre-print with @lajello: arxiv.org/abs/2312.04974
English
cocoons_dk retweetledi
📢🚨 Our study on the #TwitterMigration to #Mastodon is out now on @SciReports!
✍🏼Blogpost: communities.springernature.com/posts/get-out-…
📚Paper: nature.com/articles/s4159…
🔬The migration represents a unique example of collective behavioral change documented through large-scale digital traces.
Lucio La Cava@luciolcw
🚨 📢 Ever wondered what drives mass migration on social media platforms? 🤔 We just posted a preprint (w/ @lajello and Andrea Tagarelli) analyzing the #TwitterMigration 🐦 to #Mastodon 🐘 after Elon Musk's acquisition! 🚀 arxiv.org/abs/2305.19056
English
cocoons_dk retweetledi
Our study on the Twitter Migration to Mastodon is now out on @SciReports!
The Migration is a unique example of collective behavioral change documented through large-scale digital traces. It taught us a little about how people coordinate.
Blogpost: t.ly/zz0jA
Luca Maria Aiello (lajello.bsky.social)@lajello
The #TwitterMigration to #Mastodon is a fascinating example of collective behavioral change. We found that network structure, user commitment and language use jointly played a role in facilitating the migration. Preprint out, feedback welcome! @Carlsbergfondet @ITUkbh @nerdsitu
English
cocoons_dk retweetledi
Measuring behavior change from social media is key to understand how online interactions can promote positive practices. We surveyed the behaviors and observational techniques that researchers have used so far to capture behavior change online: arxiv.org/abs/2310.19951
English
cocoons_dk retweetledi
cocoons_dk retweetledi
A taxonomy of 217 dream topics and their shifts during COVID19 and the war in Ukraine. A large-scale NLP study on 44k dream reports from Reddit r/dreams. New preprint from our dreams research series arxiv.org/abs/2307.04167, soon at #IC2S2! #dreams #NLP #UkraineWar #covid19
English
cocoons_dk retweetledi
#IC2S2'23 BONUS EVENT 😎
"Too Lazy to Read the Paper" LIVE on July 19, 1300-1400
In this "Extravaganza" edition of his 5star⭐️ podcast series, @suneman will host 3superstars💫 of the Computational Social Science community @schichmax @clauwa and @duncanjwatts
See you there!
English