Vinson·Li

Index

On recommendation

  1. From embeddings to IDs to what you watch next

    Generative recommenders are going into production. How a video becomes an embedding, then a semantic ID, and how a Transformer uses those IDs to choose what you'll watch next.

    3 min
  2. Describe the music you want

    "Sad songs for a rainy drive" beats any genre taxonomy. Natural language is becoming the way people ask for music, and it changes what a music recommender has to understand.

    2 min
  3. Recommendation as next-token prediction

    A new paper turns every item into a short code of semantic tokens, then has a Transformer generate the code of what you'll want next. The item vocabulary finally describes what things are.

    2 min
  4. Meta's worst day was TikTok's best

    Meta lost over $230 billion in market value in one day after Facebook's first ever drop in daily users. The interest graph beat the social graph.

    2 min
  5. Listening sessions are sentences

    Treat each song as a token and each listening session as a sentence, and recommendation starts to look like language modeling. What that framing gets right, and what it misses.

    2 min
  6. You can't skim a song

    Why music recommendation is a different problem from video or news recommendation: sequences, repetition, passive listening, and the difference between a skip and a mood.

    2 min
  7. Joining YouTube Music: the recommendation is the product

    I've started as an engineering manager at YouTube Music. Why music, and why I think the next few years of music discovery will look very different.

    2 min
  8. Reels is a clone. That's fine

    Instagram launched Reels the day before the TikTok executive order. The format is a commodity now, and the question is whose recommender learns fastest.

    2 min
  9. The embedding table is the model

    Facebook open-sourced DLRM, its deep learning recommendation model. It shows what big recommenders actually look like: mostly memory, with a small network on top.

    2 min
  10. Musical.ly is now TikTok, as predicted

    ByteDance merged Musical.ly into TikTok this week. A short follow-up to what I wrote in November, and what the feed tells you about the product.

    2 min
  11. Watch the recommendation engine, not the lip-sync

    ByteDance is buying Musical.ly. It looks like a Chinese news company buying an American teen app. It's really a recommendation engine buying a global audience.

    2 min
  12. Apple bets on human DJs. I'd bet on the skip button

    Apple Music launched with Beats 1 and human curators. Three weeks later, Spotify ships Discover Weekly. Two theories of music discovery.

    2 min