To do this, ArtistInterview.ai first conducts a semantic search across the artist's lyrics for your question using OpenAI embeddings. For example, if you ask about morning routines, this search will return lyrics pertaining to breakfast even if they don't use the word “morning.”
Next, these lyric snippets are combined into a prompt that instructs GPT-3.5-turbo or GPT-4 to role play as the artist and answer your question based on the provided lyric snippets.
Even though the result isn’t actually written by the artist, it taps into the wisdom they have embedded into their lyrics, and so in a very real sense it “comes from them.”
As a co-founder of Genius.com (YC S11) I’m proud to have contributed to the last quantum leap in lyrical analysis: crowdsourcing.
Working on ArtistInterview.ai has given me a taste of the wild future of lyrical analysis, and, since music and lyrics are how we discover who we are and who we want to hang out with, it’s a pretty exciting thing.
Check it out and let me know if you have questions!
Thanks!
