In the wake of the U.S. 2024 presidential election, one fact became clear: Disinformation proliferated online at a startling rate, shaping Americans’ views about each candidate as well as a diverse set of topics, including public health, climate change, and immigration. Generative AI – with its ability to produce deepfakes in seconds and its propensity to hallucinate facts – only stands to exacerbate the problem.
Factiverse, a startup that participated in TechCrunch Disrupt Battlefield 200 in October, is bracing itself for the onslaught. The company, which won best pitch in the Security, Privacy, and Social Networking category, has developed a business-to-business tool that provides live fact-checking of text, video, and audio. The company’s pitch: to help businesses save hours of research and mitigate any reputational risk or legal liability.
The Norwegian startup is still in its early stages; Factiverse has raised around $1.45 million in pre-seed money since launching in 2020. Yet it has already begun working with both media and financial partners, including one of the largest banks in Norway, according to Factiverse CEO and co-founder Maria Amelie.
Factiverse even provided live fact-checking of the U.S. presidential debates that was used by several media partners, Amelie said.
“We’re not an LLM (large language model). We’ve built a different type of model based on information retrieval,” Amelie told TechCrunch.
As a former technology journalist and published author, Amelie has first-hand experience in the war against facts. She worked with Factiverse co-founder and CTO Vinay Setty, who is associate professor in machine learning at the University of Stavanger, to launch the startup with a B2B focus.
Factiverse’s model is trained on high quality, well-curated, and credible data from reliable sources and fact-checkers around the world, according to Amelie, and not the “junk food data” that generative AI is trained on.
“We train our AI model to intuitively think like someone who has a lot of experience with researching information,” Amelie said.
The model, which is based on machine learning and natural language processing, is able to identify claims and search the web in real time – everything from search engines like Google and Bing to AI search engines like You.com to academic papers.
“The most fun part is that we’re not showing you whatever comes up first on those search engines,” Amelie said. “We’re actually proposing to you what sources are the most, or historically have been the most, credible on your topic…We actually look into the domain in correlation to the topic, and sometimes even who is being quoted in an article.”
As of today, Factiverse says it outperforms GPT-4, Mistral 7-b, and GPT-3 in its ability to identify fact-check worthy claims in 114 languages. The company’s model also outperforms LLMs on determining the veracity of a claim. Amelie said Factiverse’s success rate is around 80% and the goal is to improve as the company onboards new customers around the world.
“We have enough funding to be the best, but we are here in the U.S. to become the fastest,” Amelie told TechCrunch. She also noted the company wants to raise a seed round in 2025. “We are looking for customers and investors who want to invest in trust and credibility,” she said.