Saloni Dash
@dashsaloni.bsky.social
740 followers 260 following 5 posts
PhD Candidate, University of Washington https://salonidash.com/
Posts Media Videos Starter Packs
Reposted by Saloni Dash
cip.uw.edu
Check out "The persuasive potential of AI-paraphrased information at scale," an open-access article exploring "AIPasta," written by a team of @cip.uw.edu researchers led by @dashsaloni.bsky.social with co-authors @yiweixu.bsky.social, @maddyjalbert.bsky.social and @emmaspiro.bsky.social ⤵️
pnasnexus.org
Repeating duplicated messages—known as CopyPasta—over and over online is a common misinformation strategy. Using AI to paraphrase the message, what the authors call “AIPasta,” could be even more effective at creating a false consensus. In PNAS Nexus: academic.oup.com/pnasnexus/ar...
#StopTheSteal AIPasta Stimuli: Profile images, usernames, and handles constructed by Jalbert et al. 2025. Profiles do not represent real users and were created from stock images and with handles that are not currently in use.
Reposted by Saloni Dash
pnasnexus.org
Repeating duplicated messages—known as CopyPasta—over and over online is a common misinformation strategy. Using AI to paraphrase the message, what the authors call “AIPasta,” could be even more effective at creating a false consensus. In PNAS Nexus: academic.oup.com/pnasnexus/ar...
#StopTheSteal AIPasta Stimuli: Profile images, usernames, and handles constructed by Jalbert et al. 2025. Profiles do not represent real users and were created from stock images and with handles that are not currently in use.
dashsaloni.bsky.social
Key Takeaway: As AI-driven information operations become increasingly common, we expect traditional CopyPasta to be replaced by AIPasta, which is harder to detect and mitigate.
dashsaloni.bsky.social
AIPasta (vs. Control) also increases belief in the exact false claim of the campaign, depending on political orientation. Current state-of-the-art AI-text detectors fail to detect AIPasta, enabling these operations to scale successfully.
dashsaloni.bsky.social
In a preregistered experiment with 1200 U.S. participants, we find that AIPasta (but not CopyPasta) is effective at increasing perceptions of consensus in false claims. Exposure to AIPasta does not change levels of sharing intent compared to a control condition (CopyPasta reduces this).
dashsaloni.bsky.social
In this paper, we study how AI-paraphrased messages can enhance information campaigns. We look at how CopyPasta - a single message crafted by adversarial actors & shared repeatedly at scale --- can be paraphrased using LLMs to generate repetitive, non-identical messages i.e., AIPasta.
Reposted by Saloni Dash
mmitchell.bsky.social
Inspired by @wimlds.bsky.social , I looked for a "Women in Machine Learning" starter pack and couldn't find one.
So I created one! May have some mistakes. I'll try to do an AI ethics one next. 🤗
go.bsky.app/LT6CwNN
Reposted by Saloni Dash
mertcanbayar.bsky.social
Great news! CIP now has a starter pack that showcases our current and past team members! Give them a follow!
Reposted by Saloni Dash
hankgreen.bsky.social
This is one of those videos that you could work on forever, but suddenly it felt very important to make it, so I finally did: www.youtube.com/watch?v=d8Pn...
Populism, Media Revolutions, and Our Terrible Moment
YouTube video by vlogbrothers
www.youtube.com