VI AI
← Back to News & Insights
Event / Lecture

VI AI Music Academy Lecture 3: Chinese Voice and Instrumental Timbre (RSI, again)

Organization / Author: VI AI Foundation · NACUAA · ZJUAANC
#Lectures #AI Music Academy

Joint community event by ZJUAANC and VI AI Music Academy.

Event Details

  • When (Three Time Zones):
    • PDT: Sat 10 Oct 2026, 5:30–6:30 PM
    • EDT: Sat 10 Oct 2026, 8:30–9:30 PM
    • Beijing Time: Sun 11 Oct 2026, 8:30–9:30 AM
  • Where: Zoom Online Meeting (Join Zoom · Meeting ID: 952 9981 2540 · Passcode: 18970521)
  • Nature: Free Community Event. Zoom. 60 minutes, including a 5-minute break.
  • Lecture Recording (YouTube): https://youtu.be/l33Ua1FHlQM
  • YouTube Series Playlist: https://www.youtube.com/playlist?list=PLGkKw_qoWmuo
  • Lecture Materials & Slides:
    • Lecture 3 Slides (Diagnostic Deck · Dark): AI Music and RSI Diagnostics / Chinese Vocals & Timbre (Download PDF)
    • Lecture 3 Slides (Audit Deck · Light): AI Vocals and RSI Audit (Download PDF)
    • Lecture 3 Slides (Reality Check Deck): AI Music and RSI Reality Check (Download PDF)
  • RSVP (This Lecture): https://luma.com/5j2b3108 (Approval required; join via Zoom upon approval)
  • Academy Website: https://music.weiai.ai

Free Zoom, 60 minutes, including a 5-minute break. Lecture 3 of 5.

First half: Chinese voice and Chinese instrumental timbre. Second half: RSI again, limited to public reports since September 2026. Not a contest-rules session.

When

  • PDT: Sat 10 Oct 2026, 5:30–6:30 PM
  • EDT: Sat 10 Oct 2026, 8:30–9:30 PM
  • Beijing: Sun 11 Oct 2026, 8:30–9:30 AM

RSVP

https://luma.com/5j2b3108
Approval required. After approval, join via Zoom.

Zoom

Clock

  • 5:30–5:55 Voice and timbre. A/B comparison: audible vs. inaudible.
  • 5:55–6:00 Break.
  • 6:00–6:25 RSI update: latest progress and reports. No model training demo.
  • 6:25–6:30 Q&A.

What We Listen For

  • Chinese articulation, vocal timbre, and breath. Does swapping the voice still hold the same song?
  • Erhu, pipa, guzheng, dizi, sheng. When prompt tags say “Chinese instruments”, do your ears actually hear them?
  • A/B version comparisons: strictly audible vs. inaudible, not personal taste.

RSI, Again

Lecture 1 already drew the line: RSI refers to a system using its own outputs to improve subsequent steps (prompts, tools, code, eval, or recipes). Bounded and layered self-improvement has been reported; a system that designs its own successor model autonomously has not.

Public claims since September 2026 (all vendor/author self-reported, not third-party audits):

  • Anthropic (internal tally to August 2026): Claude led ~26% of AI R&D tasks, collaborated in >90%, none fully autonomous. Anthropic defines full RSI as a model autonomously creating its successor, and states this has not occurred.
  • OpenAI (6 Sep 2026): claims reaching “automated research intern” milestone; targets “automated AI researcher” by March 2028. Chief Scientist discussed this timeline; automating training experiments is a separate track while humans set directions.
  • Dario Amodei (12 Sep 2026, “We Must Pace the Frontier”): argues for pacing frontier capability advances to allow time for alignment and verification.
  • Roadmap survey: arXiv:2609.11873 (submitted Sep 10, revised Sep 22), categorizing RSI into execution, strategy, experience acquisition, environment adaptation, and recursive meta-improvement.
  • Weights fixed: Dream-RSI (arXiv:2609.14858, Sep 14; Google, DeepMind, etc.), using past search records as simulators to modify search procedures without updating model weights.
  • Self-improving research agent: AIDE^2 (arXiv:2609.26457, Sep 22), claiming 7 improvements across 8 days of autonomous run, author self-reported without independent replication.
  • Anti-overfitting regularized RSI: RRSI (arXiv:2609.24972, Chinese report Oct 6), modifying prompts, control flow, tools, and memory while regularizing against task overfitting. Author-reported scores.

This lecture covers only these boundaries. Three-round song iteration remains an applied example. No model training demo.

Five-Lecture Series

  1. Sep 12: From Idea to an Editable Track, plus RSI State of the Art (Ended)
  2. Sep 26: Structure and Emotion: Chorus, Conflict, and Holding 20 Seconds (Ended)
  3. Oct 10: Chinese Voice and Instrumental Timbre (RSI, again) (Ended · Recording & Slides Available)
  4. Oct 24: Arrangement Layers and Imagery
  5. Nov 7: Final Delivery: Version Selection and Documenting Revisions

Lecture Recording & Slide Materials

Speaker

Michael Huo
President, VI AI Foundation & VI AI For Good Foundation; Founding Board Member & VP, Zhejiang University Alumni Association of North Carolina (ZJUAANC); Chair, NACUAA AI Startup Club.

Organizers

  • Organizers: ZJUAANC, NACUAA AI Startup Club
  • Organizing Hosts: VI AI Foundation, VI AI For Good Foundation
  • Co-organizers: NACUAA NC / SC / SV / NorCal / DC / WA Chapters, and Sister Zhejiang University Alumni Associations

This lecture will not discuss contest submission rules.

← Back to All Insights & Events Contact R&D Team