I skimmed a paper last night about augmenting a language model by supplying a database of facts that it uses to refine its predictions. It is meant to improve the model's ability to remember hard facts. But now I'm thinking of doing that but supplying a database of nonsense phrases or other silly things

maybe you could use that technique to quickly fine-tune a network on some chosen data instead of needing to train it 🤔

Show thread

@SuricrasiaOnline Supply one model with a collection of facts, supply another model with the negative of those, and then pit them against each other

Sign in to participate in the conversation

Cybrespace is an instance of Mastodon, a social network based on open web protocols and free, open-source software. It is decentralized like e-mail.