A Watermark for Large Language Models
Proposes a watermarking framework for proprietary language models and a statistical test for detecting the watermark using OPT.
This framework can help mitigate potential harms of large language models by embedding signals into generated text that can be algorithmically detected, providing a level of security for proprietary language models.
The Semantic Scholar Open Data Platform
Builds the Semantic Scholar Academic Graph, the largest open scientific literature graph to-date, with 200M+ papers, 80M+ authors, 550M+ paper authorship edges, and 2.4B+ citation edges.
The Semantic Scholar platform provides a robust data processing pipeline for scholarly content, offering APIs for academic use. The Semantic Scholar Academic Graph can aid researchers in their work by helping them discover and understand scientific literature faster and more efficiently than before.
Putting ChatGPT’s Medical Advice to the (Turing) Test
ChatGPT responses to patient questions were weakly distinguishable from provider responses.
This study assesses the feasibility of using AI-based chatbots for patient-provider communication, finding that laypeople appear to trust chatbots to answer lower risk health questions. This could lead to the implementation of chatbots to assist healthcare professionals in providing more efficient and effective care.