Large language models like ChatGPT have a limitation: their training is not up-to-date. If you ask ChatGPT about the effectiveness of Covid vaccines against the most common variant in 2023, it will inform you that it doesn't have access to real-time data beyond its last update in September 2021. However, the medical field constantly evolves, with new Covid strains, vaccine approvals, drug approvals, and scientific studies. To address this issue, OpenEvidence, a startup founded by Daniel Nadler, has invested $32 million and assembled a team of nearly a dozen employees with PhDs and a supercomputer in the Nevada desert. Their goal is to overcome the knowledge cutoff problem by integrating real-time clinical documents with language models, a process referred to as "retrieval augmented generation. " OpenEvidence's chatbot can provide up-to-date information by incorporating specific studies on the new Covid variant, including citations from studies published in February and May 2023. Unlike closed-book models, OpenEvidence's model can answer questions with the information available in its retrieval pool. This is not Nadler's first venture in the AI industry. He previously sold Kensho Technologies, an AI-powered tool for Wall Street traders, for $550 million in 2018. During the Covid-19 pandemic, Nadler recognized the challenge faced by healthcare providers in navigating the overwhelming amount of information.

He realized that AI could help in triaging and extracting relevant, credible information. OpenEvidence aims to compete with the incumbent database used by two million healthcare workers worldwide, UpToDate. This database relies on human experts to curate medical information, but OpenEvidence's interactive approach allows users to customize their queries and receive tailored answers. OpenEvidence retrieves information from a vast pool of over 35 million journal articles, updating it multiple times a day. The challenge lies in ensuring the quality of the sources, considering factors like impact factor rankings. While retrieval augmented systems can reduce errors, they are not infallible. OpenEvidence is currently available for free to licensed medical professionals and is contemplating a revenue model that may involve subscriptions or advertisements. However, it will not be accessible to the general public to avoid regulatory and ethical concerns, as OpenEvidence aims to support healthcare professionals and not replace their judgment.