Eric Olson is an innovator at the intersection of technology and knowledge accessibility. As the founder and CEO of Consensus, Eric has channeled his deep-seated passion for scientific exploration and expert discourse into creating a platform that democratizes access to specialized knowledge. Drawing inspiration from his own voracious appetite for non-fiction and expert-led podcasts, Eric envisioned a tool that could distill and deliver expert insights on any subject at the touch of a button. His background in technology development and a keen understanding of AI's capabilities have positioned him uniquely to tackle the challenges of unbiased information curation and user-centric design.
At the core of Consensus is a revolutionary approach to information retrieval—utilizing vector search technology to go beyond traditional keyword searches, thus capturing the nuance and interconnectedness of modern knowledge. Under Eric’s leadership, Consensus not only simplifies the academic research process but also makes it more engaging for users without a research background, encouraging a broader spectrum of users to delve into complex material with ease.
In an in-depth Q&A, Eric shares the origins of Consensus, its unique approach to leveraging AI, and his vision for the future of making expert knowledge universally accessible.
Q: Eric, can you share how the idea of Consensus originated, and what personal experiences influenced its vision to make expert knowledge accessible for all?
The concept of Consensus emerged from my passion for consuming content by scientists and experts. I enjoy reading non-fiction and listening to podcasts with scientists. I found such value in this content that I wanted a tool where I could quickly access what experts and the best evidence had to say on any topic or question I was interested in.
Q: With AI at the heart of Consensus, how do you ensure that the information remains unbiased, especially considering the complexities involved in training AI models?
The main way we avoid hallucinations and other issues with our information is through the design of our product. Consensus isn't a chatbot. We're not training a model from scratch and dealing with the problems that OpenAI or Google face with their massive language models full of data. Instead, we built a search engine with language models around it. This means we search for answers in documents, and then use AI to summarize them. We might misinterpret what's in those documents, but we won't make up information like most AI models do. We keep the information accurate by finding it first and then using AI to summarize the insights.
Q: Could you explain in layman’s terms how Consensus’ vector search technology works and how it differs from traditional keyword search methods used by other academic search engines?
Vector search adds a layer of intelligence by understanding meaning and intent, something keyword search can't do. Think of it this way: if you typed 'heart disease' in a keyword search, it wouldn't see 'cardiovascular disease' as related because it's a different word, so it would say it's not relevant. But with vector search, it knows those concepts are closely related and ranks them high. It basically adds a richer understanding of meaning and intent.
Q: How does the design of Consensus encourage users, especially those without a research background, to engage effectively with complex academic content?
Everything we do in the product aims to make it feel simple and approachable. We use color throughout to differentiate it from traditional academic tools. We have info badges on all features and tags, so users can hover over them for a simple, easy-to-understand explanation of each feature and metric. Our AI-powered synthesis features come with toggles, letting users decide to turn them on or off. This way, users can either get more information or pull back for less when they're starting out. We created tiles for papers to keep the information self-contained and straightforward: here's a paper, and here's its info. Everything in the product is designed with a low bar for user expertise, but even advanced users appreciate a simple, intuitive design.
Q: What are some of the biggest challenges you faced in the development of Consensus, and how have you overcome them? Are there any new features or technologies you are excited to integrate in the near future?
The biggest challenges for anyone building products with AI right now are that the largest models are expensive and have poor latency. This really matters for us because, in search, we need to deliver information quickly. It has forced us to get creative in using models throughout the product and often distilling lots of information into smaller models to ensure they perform well for users. As technology improves, we're excited because the models will keep getting better, even when they're smaller. This will let us handle more nuanced and complex information in a way that performs for our users. Overcoming the challenges of cost and latency with the biggest models will continue to improve as the technology advances.
Q: You mentioned that Consensus uses the Semantic Scholar database. How do you decide on additional databases or resources to include, and what criteria must they meet?
We decide on what resources and databases to use and will continue to do so, based on three criteria. First, does it fall under the umbrella of academic research? That's what our product focuses on, and that's the scope of the Semantic Scholar database. We're interested in adding more academic research resources over the next year or two. Second, what's the cost and effort involved in integrating that resource? Semantic Scholar has clean, high-quality data, and partnering with them was a no-brainer for us. Cost and effort will always be a factor in deciding future partnerships. Third, what's the quality of the content? What journals are included, and what's the database's coverage? Semantic Scholar works with some of the top journals, and we'll always consider the quality and coverage of any new database before adding it.
Q: Given your background and the educational focus of Consensus, are there any partnerships with educational institutions or plans to integrate Consensus into academic curricula?
We're already used by students and faculty at over 5,000 universities and schools around the world, and we're excited to keep growing. Many of them started using our product on their own, but in the future, there's plenty of opportunity for us to work directly with schools or faculty where they pay for access, providing it to many students at once. If we can be integrated into faculty courses on how to use research in your work, we'd love to explore those opportunities.
Q: How does Consensus handle feedback and community interaction to improve its search algorithms and user experience?
User feedback is incredibly important to us at Consensus. We're constantly conducting user interviews and talking to our users. We have a Slack channel with them where we're always getting feedback. We also have a very interactive support portal where people share complaints or new ideas. My co-founder and I check it daily, responding and talking to users. Our entire roadmap of what to build next is dictated by what our users tell us. We're a super user-guided company, and everything we decide to build is informed by the problems our users face with our product and others, which we aim to solve.
Q: In what ways does Consensus address ethical concerns related to AI and data privacy, particularly when handling sensitive or proprietary research data?
All user data we store is anonymized, and we don't share it with any third parties. It's all kept internally. We're not training models from the ground up, so nothing our users do is being used to train AI models as it would be with ChatGPT or OpenAI. Everything is private, and we don't give anything to third parties.
Q: Looking forward, how do you see Consensus evolving in the next five to ten years? What role do you envision it playing in the broader landscape of academic research and public knowledge consumption?
Our goal with Consensus is to make the best information in the world more accessible and consumable. Academic and scientific research is the natural starting point, and we have a long way to go there. So, for the next few years, we'll be laser-focused on that domain and building the best search and analysis tool for academic and scientific research. Once we've made progress on that front, we'll start expanding into other datasets. The world's best knowledge isn't just in academic research. It’s a great starting point, but there are other datasets with excellent curated information. We could explore healthcare, market research reports, or public data from the government. We want to eventually expand Consensus into a go-to source for anyone seeking authoritative information and deep research.
Comments