Latest research
November 2023 - Does GPT-4 have theory of mind capabilities?
The FANToM benchmark analyzes theory of mind capabilities of 13 state-of-the-art LLMs based on essential criteria.
November 2023 - Are you thirsty for social chitchat data?
We introduce SODA, the first million-scale high-quality social chitchat dataset.
November 2023 - SPECTER2: Adapting scientific document embeddings to multiple fields and task formats
We create SPECTER2, a new scientific document embedding model via a 2-step training process on large datasets.
November 2023 - Faith and fate: Limits of Transformers on compositionality
Why and when do Transformers succeed or struggle in compositional problems? Find out in our Neurips 2023 paper.
October 2023 - Case study: Iterative design for skimming support
The Semantic Scholar team announces a skimming feature to help researchers keep up with the latest publications.
September 2023 - Ai2’s Wildlands team collaborates with FireWatch to increase community fire resilience
The Wildlands and FireWatch are assisting homeowners and firefighters in preventing the spread of wildfires.
August 2023 - Ai2 Dolma: 3 trillion token open corpus for language model pretraining
We introduce Dolma, an open dataset from web content, academic publications, code, books, and encyclopedic…
February 2023 - The artist and AI: Familiar strangers
Many artists are skeptical about AI. But is AI really the monster, or a misunderstood artist’s assistant?
January 2023 - America needs AI literacy now
In late 2021, we conducted a survey of 1,547 US adults, and found that most Americans are ill-informed about AI.