Join top executives in San Francisco on July 11-12, to hear how leaders are integrating and optimizing AI investments for success. Learn More
Boston-based Consensus, an AI-powered search engine aimed at scientific research, announced today that it has secured $3 million in a seed funding round to continue its mission to improve scientific web search quality.
Consensus’s search engine aims to solve the problem of biased and inaccurate search results by delivering expert knowledge from 200 million scientific and academic research papers. With nearly 200,000 registered users since its launch in September, the platform aims to offer genuine answers to life’s most interesting questions.
Courtesy of a co-op with OpenAI, the search engine leverages the most recent, relevant and authoritative sources to provide plain-language summaries of results, addressing one of the most significant problems in searching for expert information.
>>Don’t miss our newest special issue: Data centers in 2023: How to do more with less.<<
Join us in San Francisco on July 11-12, where top executives will share how they have integrated and optimized AI investments for success and avoided common pitfalls.
CEO Eric Olson and co-founder Christian Salem, who come from an academic background, created the Consensus app with total funding of $4.25 million, including reinvestment from pre-seed round backer Winklevoss Capital.
“We value the truth and have always wanted an easy way to engage with the rigorous source material,” Olson told VentureBeat. “We built Consensus because we wanted it to exist for ourselves.”
With the endorsement of Tim Draper of Draper Associates, which led the seed funding round, the company aims to revolutionize scientific web search, transform research and disrupt the $200 billion global industry.
“We believe that the way Consensus transforms research is just the beginning of a sea change in how web users obtain information,” said Draper. “Consensus will own Search 3.0.”
Eliminating bias to serve true search results
Consensus distinguishes itself from popular counterparts by prioritizing sources’ authority rather than popularity and user preferences. This approach prevents the click-baity, SEO-hacked results that often mislead users with bias and misinformation. The platform’s cutting-edge generative AI technology synthesizes answers directly from its pool of over 200 million research papers.
“Our goal is to use language models to automate the steps that an expert would take in rigorously concluding a question, and build an intuitive and easy-to-use search experience around those automated steps,” Olson told VentureBeat.
A customized version of GPT-4 present in the architecture’s backend, developed in partnership with OpenAI, generates a plain-language summary of Consensus’s results, providing users with reliable, evidence-based information.
“Through our relationship with OpenAI, we were able to get early access to the customized GPT-4 API,” explained Olson. “We were internally working on our summary feature for months and shipped the new feature using GPT-4 in just five days after getting access to it.”
Olson said that the search engine’s proprietary “claim extractor” identifies word-for-word claims by paper authors. The generative AI model provides users with a list of the 10 most relevant claims to their queries and then employs GPT-4 to generate a clear and concise summary of those top 10.
With this technology, users can quickly and easily navigate large volumes of information, helping them to make informed decisions based on relevant and accurate data.
The search engine’s innovative “Consensus Meter” also provides a percentage-based assessment of the veracity of answers to yes-or-no questions, saving users time and effort when searching for a simple answer.
“For yes-or-no questions, we look at the top 20 results and classify which side of the fence they sit on (i.e., do they indicate the answer to your question is yes, no, possible, or other),” said Olson. “Once that is complete, we show you a percent count of each ‘school of thought’ in the Consensus Meter.”
Revamping the search engine user experience
The boom in generative AI and large language models (LLMs) has led to a proliferation of AI-powered scientific search engines. These tools aim to simplify researchers’ access to scientific papers and summarize the major findings in a particular field. With so many developers claiming that their app will democratize and streamline access to research, it can be challenging to determine which search engine is the most effective.
Scientific AI search engine Elicit, which uses an LLM to craft its answers, searches papers in the Semantic Scholar database and identifies the top studies by comparing the papers’ titles and abstracts with the search question. Another tool, scite.ai, which claims to be “ChatGPT for science,” uses an LLM to organize and add context to paper citations — including where, when and how another paper cites a paper.
Setting Consensus apart
According to Consensus, legacy expert/academic search engines like Google Scholar haven’t innovated in decades, and the scientific AI search engines currently available provide an unsatisfactory user experience.
“We are taking the latest and greatest technology being used on general-purpose consumer tools like ChatGPT and applying them thoughtfully and intentionally in a specific vertical like research,” said Olson. “Our product is smarter, more efficient and generally delivers a better user experience.”
He added that unlike others, the Consensus app does not require exact keyword matching when users input a query; they can ask a plain-English, natural-language question.
“Our AI models actually pull out answers from papers to your question and do not just deliver back a list of blue links,” he said. “We also provide quality indicators about each paper beyond just citation count, like the journal quality and type of study design.”
A future of opportunities with generative AI
Olson told VentureBeat that the company aims to become the go-to search product for expert information. However, unlike broad generative models like ChatGPT, Consensus is focused on answering consequential questions that require expert opinions.
“Scientific research was a natural starting point, but eventually we want to expand our dataset to other places where expert knowledge exists in large text datasets, like market research or financial reports,” he said.
With its recent funding, the company plans to further develop its generative AI technology and expand its user base. Its mission is to improve the quality of consequential web searches and provide reliable, evidence-based information to all users, regardless of their expertise.
“We are going to double the size of our engineering team to enable us to move faster to deliver all of the amazing features that our users are asking us for,” said Olson. “Whether people look up information at work, at school or just in an argument with their friends, we want Consensus to be synonymous with unbiased, factual information.”
The oversubscribed $3 million seed funding round was led by Draper Associates, including celebrated seed-stage investor Tim Draper, whose previous investments include Tesla, SpaceX and Baidu. Additional investors in the round include Kevin Carter (Crunchbase, Snap, Pinterest), Brian Pokorny (Twitter, Square, Stitcher), Nomad Capital (Open Sea, Intercom) and members of the OpenAI research team.
VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.