Proem: AI-Driven Platform Democratizes Science by Summarizing Research for All
It’s an open secret that scholarly articles are reviewed and kept on a preprint server before they are published. But while these articles are publicly available, they are “not easily digestible,” argued Geet Khosla, a serial tech entrepreneur and CEO of Proemial, who gave Machine Design a first look at his company’s open and free platform, proem, which aims to democratize scientific knowledge by providing access to more than 240 million research papers globally.
Not only does proem use large language models to summarize and explain complex research topics, but the platform also offers customized feedback so that users can easily cull up-to-the-minute insights and breakthroughs directly from a repository of scientific research. Proem’s algorithms extract a short summary, allowing users to browse quickly through available papers on a given topic.
The platform is the brainchild of Khosla and his co-founders Mads Rydahl, the first head of product and design at Siri and Brian Pedersen, AI product visionary and tech veteran, hacker and an experienced CTO.
READ MORE: Aveva World Conference 2024: Designing for a Connected Industrial Economy
According to Khosla, proem has two standout features: The platform adds more than 410,000 new research papers every month and was designed to allow anyone to delve into specific scientific papers.
Balancing Peer Review and Fast Access to Scientific Insights
The fields of AI, computer science, robotics and STEM, in general, are moving faster than the speed with which humans can consume and process information. “You don’t want to read a one-year-old newspaper because it’s old news and because the field has moved on,” said Khosla, who is based in Aarhus, Denmark.
The laborious task of reading peer-reviewed journals, specifically in fast-moving technical fields, has become outdated, claimed Khosla. “What we care about is giving access to those who don’t care about peer review because their job isn’t to argue about the specifics of the study,” said Khosla. “Their job is to take the insights from research and apply them to their job, to improve or to better connect with people.”
Creating an Accessible Digital Commons for Scholarly Research
Proemial’s search engine is configured to communicate in conversational, non-technical language, making it accessible to individuals with limited scientific knowledge. Proem also adapts to users’ language style, said Khosla.
“The problem with research is that it is jargon-heavy and it is specifically focused on a very niche area,” he explained. “The niche area is not a problem, but the jargon is a problem. And we want to essentially change that by not just making it a ChatGPT-like experience, where you’re talking to a model and getting an answer without any references, but we’re also trying to build a new way for teams to collaborate.”
Since users can ask questions and leave their comments on the platform, they are able to build on the topic collaboratively—either openly or privately in teams when companies want to pay for private access.
One of Proemial’s partners is a large R&D company that employs about 40,000 researchers who no longer work in academia. “Their job is to take the research and apply it to a commercial problem,” Khosla said. “And that’s where we want to come in. We don’t believe the insights finish on a paper. The insights start on a paper.”
READ MORE: Q&A: Upskilling Manufacturing Factory Workers With AI
Proemial’s waitlist includes individuals from organizations like CERN, Novo Nordisk, Google (Alphabet), Facebook (Meta) and Amazon, as well as academic institutions like MIT and Cornell.
Addressing Accuracy and AI Hallucination in Proem’s Platform
Proemial ensures that every answer and every piece of content is based on research, said Khosla. “Nothing on the platform is ever purely the model,” he said. “It’s always the model confined to all the research—multiple papers or one paper. We do sometimes have the model putting an output without a connection, but that’s just to get you to a paper.”
The goal, said Khosla, is to always base the output of the model on research. “That way we avoid the hallucination problem because you never talk directly to a model alone. You always get the model with the backing of references and research.”
Click here to test the research tool.
READ MORE: Epicor Introduces AI and BI Capabilities with its Grow Portfolio