Lex Fridman Podcast Gustav Soderstrom: Spotify
7 snips
Jul 29, 2019 
Gustav Soderstrom, Spotify's Chief R&D Officer, shares insights on how machine learning enhances music personalization and user experience. He delves into the emotional and social aspects of music consumption, emphasizing the shift from physical media to digital streaming. Gustav also discusses Spotify's role in revolutionizing podcasting, blending music and audio for an enriched user journey. The conversation touches on the balance between human curation and algorithms, highlighting the future of audio technology and its impact on creativity.
AI Snips
Chapters
Transcript
Episode notes
Building Taste Vectors
- Building user taste vectors is challenging due to the dynamic and nuanced nature of preferences.
- Deep learning embeddings help uncover latent patterns and capture cultural influences.
Contextual Preferences
- Song preferences are contextual, influenced by mood, activity, and environment.
- Spotify uses fault-tolerant UIs to accommodate the limitations of current machine learning.
User Feedback Signals
- Use play-through rate, saves, playlists, and explicit likes as signals for user preferences.
- Weigh these signals differently due to varying noise levels: high volume vs high trust.