NSF-SciFy Presents 2.8M Scientific Claims Dataset at ACL 2026

Source Posts on X

Cluster sources

📢 I am incredibly proud of NSF-SciFy that we are presenting at #ACL2026 today! What? 𝟚.𝟠𝕄 𝕤𝕔𝕚𝕖𝕟𝕥𝕚𝕗𝕚𝕔 𝕔𝕝𝕒𝕚𝕞𝕤 𝕗𝕣𝕠𝕞 𝟜𝟘𝟘𝕂 ℕ𝕊𝔽 𝕒𝕨𝕒𝕣𝕕 𝕒𝕓𝕤𝕥𝕣𝕒𝕔𝕥𝕤, 𝕤𝕡𝕒𝕟𝕟𝕚𝕟𝕘 𝕕𝕚𝕧𝕖𝕣𝕤𝕖 𝕊𝕋𝔼𝕄 𝕒𝕣𝕖𝕒𝕤! 🧵 https://x.com/deliprao/status/2074486210811089161/photo/1

Delip Rao e/σ@delipraoTECH

This largest (2.8M) scientific claim dataset ever from a source never-explored before — NSF award abstracts spanning multiple decades and disciplines! For context: the largest prior dataset had ~12K claims. https://x.com/deliprao/status/2074486227412074837/photo/1

Delip Rao e/σ@delipraoTECH

Why is this important? Remember LK-99 frenzy? Or black plastic spatula scare? How do we verify claims like that? It starts with having a high quality scientific claims dataset. https://x.com/deliprao/status/2074486247242756354/photo/1

Delip Rao e/σ@delipraoTECH

Why now? With rise of AI scientists, there is a hunger for solving non-verifiable problems, and scientific claim verifiers are a key part of that. This dataset helps you build such verifiers in multiple areas of science.

View all

Delip Rao e/σ@delipraoTECH

This open dataset is also a source of a new category of NLP problems — investigation proposals (what researchers promise to study) as opposed to claims (what they assert is true).

Delip Rao e/σ