IS THIS INFO VALID?

Is This Info Valid? was created to deal with the broken filter bubble. At some point, while the 2024 election cycle started heating up and the family group chat had already forwarded a bot-generated image as breaking news, it became clear that the rule "If it's coming from a credible channel, it's probably credible" didn't work anymore.

The statistics confirm this. NewsGuard has been tracking the increase of AI-generated "news" websites from 49 in February 2023 to over a thousand as of the spring of 2024, and most of them publish articles based on lies with zero human editing involved. MIT researchers discovered that false claims travel through social media about six times faster than real ones in 2024, and according to Pew Research Center data, the vast majority of U.S. adults regularly run into news that appears to be generated by artificial intelligence.

Of course, there is something more alarming than fake news. In general, large language models tend to hallucinate. They will create a fictional court case, invent a number, generate a quote from an actual person without any connection to reality, and all that while using the same tone of voice for all the information, even if the sentence is obviously false. Thus, when a chatbot makes a mistake, it tends to make it in a sophisticated manner, which makes it hard to spot immediately.

This is why I needed a tool to do what once did not need doing. Break down a single sentence, identify all the sources of each element, and then classify the sources depending on how much effort went into their creation: if the source is a primary document, wire reports, peer-reviewed studies, or someone's vibe check, respectively. Not to evaluate it and say either "false" or "true," but simply analyze it as an intelligent being would do.

That way, if at least a couple of users would think twice before resharing a post and would bother checking the list of citations behind the story, it will be a great accomplishment. The rest will take care of itself.