Training Data Suno's Mikey Shulman: Everyone Can Make Music Now
10 snips
May 13, 2026 
Mikey Shulman, founder and CEO of Suno, builds AI tools to make music creation accessible to everyone. He talks about modeling raw audio waves and choosing autoregression to generate full songs. He explores genre-bending sounds, why creation is the entertainment, partnerships with major labels, and visions for interactive live shows and social co-creation.
AI Snips
Chapters
Transcript
Episode notes
Discord Jam Sessions Validated The Idea
- Suno launched a Discord bot early to validate delight despite poor initial audio quality.
- Early users loved making short clips and that feedback convinced the team to keep building a consumer music product.
Modeling Sound Waves Unlocks New Music
- Suno models raw audio as continuous waveforms instead of encoding musical rules, enabling unlimited tones and novel sounds.
- Mikey describes sampling at 48,000Hz and treating audio as float32, which unlocked genre-blending and microtonal outputs.
Music Relies On Preference Data Not Pure Scale
- Music isn't primarily a scale-driven problem like LLMs; models stay relatively small and human preference data matters more.
- Suno uses user preference signals and research cycles to iteratively align models to taste rather than relying on sheer compute.