We detect and verify checkworthy statements in 114 languages, faster than any LLM, for precise, high-stakes decisions. NATO DIANA 2025 / CEO of @FactiverseAS
1/5 - Our newest findings.
Analysing 1 mill sentences with @claudeai or @GeminiApp: 483-833 hours.
With @FactiverseA : 22 minutes.
What do we mean by analysing?
1/5 - Our newest findings.
Analysing 1 mill sentences with @claudeai or @GeminiApp: 483-833 hours.
With @FactiverseA : 22 minutes.
What do we mean by analysing?
Worth a read: Estonian Language Institute (ELI) with help of @propastop and numerable experts ranked dozens of LLMs on their ability to avoid “taking positions on topics that the Russian Federation uses in its strategic narratives.”
They also developed a Propaganda Resistance benchmark - check out who is worst.
Reported in
@arstechnica
https://t.co/bI9bUY0wfa
https://t.co/6qESYorpSE
"I analysed 30 hours of US midterms videos in 1 hour yesterday before dinner." My colleague Seán Jacob wrote to me on Slack.
With Factiverse, he extracted 1,346 actual check-worthy statements in 1 hour. Who said what about whom, and whether any of it holds up.
If Seán had used just an LLM for this, it would have taken him hours, cost a fortune, and been less accurate.
We keep beating the frontier LLMs specifically on claim detection in 114 languages.
We're the best in the world at one specific thing: claim detection.
Give us a million sentences, and we find the ones worth checking.
"Maria likes penguins" → opinion, discard.
"In 2026, Maria put €5M into a penguin-welfare company" → a check-worthy factual claim that we detect and verify.
We do this in 114 languages, better than any frontier LLM today. Ten years of @vinaysetty 's peer-reviewed research underneath it.
This is the eval signal that's missing when you train a model on what it says vs.
More benchmarking data soon - tune in.
Tonight @FactiverseAS has successfully performed AI live fact-checking on the US Presidential debate between Vice President Kamala Harris and Former President Donald Trump.
We detected 301 claims made by both candidates; 456 claims fewer than during the first debate.
@FactiverseAS Factiverse are partners in Nordis, an EU-funded project supporting Nordic fact-checkers. Based on feedback from the Nordic media community, we piloted the live fact-checking feature. We are finalizing the findings from the debate. Reach out if you are want to learn more.
Glimrende kritikk av saken som ble ukritisk publisert i @vgnett i går. Det er ikke noe som heter norsk is! Den består av råvarer fra Madagaskar, Nigeria, Italia og Spania. Faktasjekk politikere, og hør på @andersgundersen i @sensonomic
https://t.co/wTe4IYqBfz