Nadav Borenstein

Which, whose, and how much knowledge do LLMs represent?

I'm excited to share our preprint answering these questions:

"Epistemic Diversity and Knowledge Collapse in Large Language Models"

📄Paper: arxiv.org/pdf/2510.04226
💻Code: github.com/dwright37/ll...

1/10

13.10.2025 11:25 — 👍 89    🔁 26    💬 2    📌 1

Qualtrics Survey | Qualtrics Experience Management The most powerful, simple and trusted way to gather experience data. Start your journey to experience management and try a free account today.

Crowdsourcing datasets is very common, but how much of those results come from LLMs these days? How can we prevent it? Much of it is unclear. This is why we’re conducing a survey (~10 min) to gather community experiences, challenges, and solutions.
Share your thoughts 👉https://tinyurl.com/39ab55wf

05.09.2025 11:32 — 👍 3    🔁 3    💬 0    📌 1

If you are attending #ACL2025, come and check out our cool poster on Wednesday at 11 am!

(Oh yeah, I forgot to announce that the paper got accepted to the main proceedings of #ACL2025, oops)

28.07.2025 20:26 — 👍 12    🔁 3    💬 0    📌 1

I am happy to share that the paper got accepted to the main proceedings of ACL 2025 in Vienna! See you there!

arxiv.org/abs/2502.14132

28.05.2025 06:42 — 👍 9    🔁 1    💬 0    📌 0

Massive congrats to @nadavb.bsky.social for passing his PhD defence yesterday! 👏👏👏
Thanks to Anders Søgaard, @radamihalcea.bsky.social @ricosennrich.bsky.social for serving as examiners.
You can find his thesis here, which the committee characterised as "a joy to read": arxiv.org/abs/2503.07395

20.03.2025 15:00 — 👍 29    🔁 3    💬 0    📌 0

This, in turn, will limit the efficacy of community notes, particularly for high-stakes claims related to health and broader conspiracy narratives.

Work done with @gretawarren.bsky.social, @delliott.bsky.social, and @iaugenstein.bsky.social.

21.02.2025 10:29 — 👍 2    🔁 0    💬 0    📌 0

Fig. 3. strategies in debunking claims related to broader nar- ratives. (b) the different ways in which fact-checking sources are used to debunk claims. Fig. 4. Distribution of notes’ topics, with and without a fact- checking source. Fig. 5. Percentage of samples related to a broader narrative or conspiracy vs. have a fact-checking source.

Fact-checkers conduct in-depth research beyond the reach of amateur users, while community notes publicise their work. The move to end partnerships and funding for fact-checking agencies will hinder their ability to pursue investigative journalism, which community note writers rely on.

21.02.2025 10:29 — 👍 3    🔁 0    💬 1    📌 0

This work examines how this recent development might impact the efforts to combat misinformation spread on social media by studying fact-checking’s role in Twitter’s community notes.

We find that community notes and professional fact-checking are deeply interconnected:

21.02.2025 10:29 — 👍 2    🔁 0    💬 1    📌 0

The current political climate has prompted Meta to shift away from partnerships with fact-checkers to increased reliance on crowdsourced community notes—the model used by Twitter.

21.02.2025 10:29 — 👍 2    🔁 0    💬 1    📌 0

Can Community Notes Replace Professional Fact-Checkers? Work done by Nadav Borenstein, Greta Warren, Desmond Elliott, and Isabelle Augenstein.

🚨 Preprint alert 🚨

𝐂𝐚𝐧 𝐂𝐨𝐦𝐦𝐮𝐧𝐢𝐭𝐲 𝐍𝐨𝐭𝐞𝐬 𝐑𝐞𝐩𝐥𝐚𝐜𝐞 𝐏𝐫𝐨𝐟𝐞𝐬𝐬𝐢𝐨𝐧𝐚𝐥 𝐅𝐚𝐜𝐭-𝐂𝐡𝐞𝐜𝐤𝐞𝐫𝐬?

(arxiv.org/abs/2502.14132)

Fact-checking agencies have come under intense scrutiny in recent months regarding their role in combating misinformation on social media.

21.02.2025 10:29 — 👍 12    🔁 7    💬 2    📌 3

Hello, world! We're a Natural Language Understanding group at the University of Copenhagen (faculty: @iaugenstein.bsky.social, @apepa.bsky.social). Follow us for updates about our research, open positions, and more!
More about us: www.copenlu.com
#NLP #NLProc #ML #XAI

20.11.2024 14:45 — 👍 14    🔁 3    💬 1    📌 1