VI AI Music Academy Lecture 3: Chinese Voice and Instrumental Timbre (RSI, again)
Forwarded by NACUAA Northern California Chapter.
Event Time & Venue
Free Community Event. Zoom. 60 minutes, including a 5-minute break.
VI AI Music Academy Season 1 (5 Lectures). This is Lecture 3.
First half: Chinese voice and Chinese instrumental timbre. Second half: RSI again, limited to public reports since September 2026. Not a contest-rules session.
Not a contest-rules session.
Schedule Clock
- 5:30–5:55 Chinese voice and instrumental timbre. A/B comparison: audible vs. inaudible.
- 5:55–6:00 Break (5 minutes)
- 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. This is a survey, not a live demo.
- 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. Public notes indicate code not yet released.
- 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 Slides & Recording (Dual Mirror Channels)
PDF Format · Free DownloadLecture 3 Full Recording (YouTube)
Full 60-minute presentation & live listening test with Michael Huo
Core Slides 1:《AI Music and RSI Diagnostics》（Diagnostic Deck · Dark）
Chinese Vocals & Timbre, Traditional Instruments A/B Tests, 4 RSI Papers Breakdown & Frontier Boundaries
Core Slides 2:《AI Vocals and RSI Audit》（Technical Audit Deck · Light）
AI Vocals & Timbre Audit, RSI Horizon & Frontier Research Verification
Core Slides 3:《AI Music and RSI Reality Check》（Reality Check Deck）
Objective Signal vs. Tech Bubble: Human Aesthetics as Mentor for Iteration, Autonomous Progress Audit
Speaker
- • 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)
Related Links & Recording
- • Recording (YouTube): https://youtu.be/l33Ua1FHlQM
- • Series Playlist (YouTube): https://www.youtube.com/playlist?list=PLGkKw_qoWmuo
- • Academy Website: https://music.weiai.ai
- • Lecture 3 Luma: https://luma.com/5j2b3108
- • Lecture 1 Luma: https://luma.com/d96kvs6t
- • Lecture 2 Luma: https://luma.com/z0329spm
Free community event. No payment or verification code required.
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.