Neural FM
A hackathon project that let you control your own information timeline by training neural-network classifiers on social media posts and news stories.
I built Neural FM for the Bitcoin SV Hackathon in the summer of 2020. Each channel was a feed of posts and news stories filtered by a small neural network. You voted on stories, retrained the model in a click, and the feed reshuffled to match. You could fork another person’s channel and change only the parts you wanted: the data source, the feature extractor, the normalizer or the network. Every model and every editorial decision was recorded on Bitcoin, so anyone could audit why a story was shown.
The idea of scoring the news with a model, instead of letting a platform decide, came back three years later in News Score.