AI summary by TwinMind
Did AI Just End Human Made Music? Ft. Rick Beato
ColdFusion
25m 46s
Introduction and AI Music Demonstration
- Video sponsored by Brilliant learning platform
- Cold Fusion episode on AI-generated music revolution
- Multiple AI-generated song examples demonstrated:
- 100% AI-generated theme song (lyrics via ChatGPT)
- Classical piece, guitar-driven rock, R&B, UK garage
- Album covers also AI-generated
- Host: musician for 20 years, producer for decade
- Mixed feelings about AI impact on personal music creation
AI Music Platform Overview
- Two main platforms: Udio and Suno AI
- Both allow text-prompt music generation without musical knowledge
- Udio: 30-second tracks, created by ex-Google DeepMind engineers
- Suno: 2-minute tracks, up to version 3, Microsoft co-pilot partnership
- Both platforms free to use currently
- Users input subjects, instruments, feelings, custom lyrics
- Tracks ready in 1-2 minutes with extension/variation options
Traditional Music Production vs AI
- Traditional process demonstrated with acoustic guitar riff example:
- Recording with Gretsch electric guitar, delay, reverb, distortion pedals
- Building bass, drums, atmosphere in Ableton
- EQing, compressing, adding effects
- Solving frequency clash issues
- Adding vocals and arrangement
- Total time: days to weeks per song
- AI process: type words, get complete song instantly
Udio Technical Observations
- Advanced production techniques included:
- Side chaining, tape effects, reverb, delay on vocals
- Realistic vocal harmonies
- Electric guitar pickup sound mimicking
- Weaknesses identified:
- Occasional major failures in output
- Limited flexibility once generated
- Low fidelity on some tracks
- Struggles with UK jungle genre
Company Backgrounds
- Suno AI: Founded late 2023, 4 founders from Kensho (financial data AI startup acquired by S&P Global)
- Udio: Founded late 2023, recently public, ex-Google DeepMind researchers
- Backing: Andreessen Horowitz VC, musicians Common and Will.i.am
Historical Context of Computer Music
- 1957: First computer-aided composition "Illiac Suite" by Lejaren Hiller and Leonard Isaacson
- 1984: David Cope's Emmy (Experiments in Musical Intelligence)
- Generated music in styles of different composers
- Database-driven approach analyzing 371 Bach chorales
- 1985: Computer accompaniment system with pattern matching
- 1990s: David Bowie's Verbasizer for lyrical writing
- 2012: Shift from hard-coded algorithms to neural networks
Modern AI Music Evolution
- 2016: Google Project Magenta piano piece
- Google MusicLM: Text-to-music generation
- OpenAI Jukebox, Stable Audio, Adobe Music Gen, YouTube Dream Track
- Previous limitations: rigid, non-human sounding compositions
- 2017: Taran Southern - first commercial AI music release
- Used 4 different AI technologies
- Required significant manual work vs. current text-prompt systems
Technical Implementation
- Based on neural networks processing vast data patterns
- More complex than text generation due to multiple variables:
- Instrument tone, tempo, rhythm, sound design
- Volume, compression, EQ considerations
- Genre understanding and coherent output
- Audio diffusion technology:
- Adds/removes noise from signal until desired output achieved
- Similar to image diffusion starting with random noise
Copyright and Legal Concerns
- OpenAI CTO Mira Morati questioned about Sora training data
- User investigation: ABBA "Dancing Queen" similarity test
- Input similar lyrics without mentioning band/song
- Output very close to original melody, rhythm, vocal cadence
- 200+ artists signed open letter against AI infringement
- Signatories: Bon Jovi, Billie Eilish, Metro Boomin
- Suno investor Antonio Rodriguez aware of potential lawsuits but invested anyway
- Companies claim regular conversations with record labels
Industry Impact Analysis
- AI music compared to "printing money" causing inflation
- Music industry already challenging for non-top 1% musicians
- TikTok transformation of music consumption
- Digital platform relationships with musicians problematic
- AI creates "unlimited supply" tsunami effect
- Stock/royalty-free music creators most at risk
Rick Beato Interview Insights
- Won't completely replace musicians - people enjoy playing instruments
- Will impact artist income through market dilution
- Predicts AI Billboard #1 hit within 2 years
- Expects 9/10 top songs AI-generated within 10 years
- People won't care if AI-generated if they like the song
- Excited about AI for: mastering, mixing, vocal editing, rough video edits
- Comparison to 1980s drum machine adoption
- Current music already heavily synthetic (samples, autotune, pitch correction)
Future Implications and Concerns
- 2023: US federal judge ruled AI artwork can't be copyrighted
- Need for new copyright legislation around AI and art
- Live human music will become increasingly valuable
- Risk of "AI fatigue" diminishing appreciation for human-made music
- Author's personal conflict: loss of human creative journey vs. democratization
Historical Parallel: 1997 Bach Competition
- University of Oregon battle: Dr. Steven Larson vs. Emmy computer
- Three pieces performed: original Bach, Larson composition, Emmy generation
- Audience incorrectly identified Larson's as computer-made
- Thought Emmy's composition was original Bach
- Dr. Larson's distress: "very disconcerting" people could be "duped"
- David Cope: AI can make "beautiful music but maybe not profound music"
- Human messiness in art creation is what we relate to most
