Within the wake of the U.S. 2024 presidential election, one truth grew to become clear: Disinformation proliferated on-line at a startling fee, shaping Individuals’ views about every candidate in addition to a various set of subjects, together with public well being, local weather change, and immigration. Generative AI – with its potential to supply deepfakes in seconds and its propensity to hallucinate info – solely stands to exacerbate the issue.
Factiverse, a startup that participated in TechCrunch Disrupt Battlefield 200 in October, is bracing itself for the onslaught. The corporate, which gained greatest pitch within the Safety, Privateness, and Social Networking class, has developed a business-to-business device that gives reside fact-checking of textual content, video, and audio. The corporate’s pitch: to assist companies save hours of analysis and mitigate any reputational threat or authorized legal responsibility.
The Norwegian startup remains to be in its early phases; Factiverse has raised round $1.45 million in pre-seed cash since launching in 2020. But it has already begun working with each media and monetary companions, together with one of many largest banks in Norway, in line with Factiverse CEO and co-founder Maria Amelie.
Factiverse even offered reside fact-checking of the U.S. presidential debates that was utilized by a number of media companions, Amelie mentioned.
“We’re not an LLM (giant language mannequin). We’ve constructed a distinct sort of mannequin based mostly on info retrieval,” Amelie informed TechCrunch.
As a former expertise journalist and revealed writer, Amelie has first-hand expertise within the warfare in opposition to info. She labored with Factiverse co-founder and CTO Vinay Setty, who’s affiliate professor in machine studying on the College of Stavanger, to launch the startup with a B2B focus.
Factiverse’s mannequin is educated on prime quality, well-curated, and credible information from dependable sources and fact-checkers around the globe, in line with Amelie, and never the “junk meals information” that generative AI is educated on.
“We prepare our AI mannequin to intuitively suppose like somebody who has lots of expertise with researching info,” Amelie mentioned.
The mannequin, which relies on machine studying and pure language processing, is ready to establish claims and search the online in actual time – every thing from search engines like google and yahoo like Google and Bing to AI search engines like google and yahoo like You.com to tutorial papers.
“Essentially the most enjoyable half is that we’re not displaying you no matter comes up first on these search engines like google and yahoo,” Amelie mentioned. “We’re truly proposing to you what sources are essentially the most, or traditionally have been essentially the most, credible in your subject…We truly look into the area in correlation to the subject, and generally even who’s being quoted in an article.”
As of immediately, Factiverse says it outperforms GPT-4, Mistral 7-b, and GPT-3 in its potential to establish fact-check worthy claims in 114 languages. The corporate’s mannequin additionally outperforms LLMs on figuring out the veracity of a declare. Amelie mentioned Factiverse’s success fee is round 80% and the purpose is to enhance as the corporate onboards new prospects around the globe.
“Now we have sufficient funding to be one of the best, however we’re right here within the U.S. to turn out to be the quickest,” Amelie informed TechCrunch. She additionally famous the corporate needs to boost a seed spherical in 2025. “We’re on the lookout for prospects and buyers who wish to put money into belief and credibility,” she mentioned.
