Unique Research · Insight / Unique Friend Q&A · January 16, 2024 · Guest: Jiang Yuchen, Founder, AIWaves / Waveform Intelligent (波形智能)
Inventory type remains insight (Unique Friend series — not partner welcome).
Guest
Jiang Yuchen — Founder, AIWaves Inc. (Hangzhou). ZJU Chu Kochen College advanced engineering; PhD CS, ETH Zurich under Prof. Ryan Cotterell (ACL Rising Star) — controllable NLG, long-form, MT, coreference. Papers/reviews at ACL, ICML, EMNLP, NAACL; WMT Shared Task organizer; MSRA work on LLM training/deploy (incl. ChatGPT-class).
Mission: AGI that improves human life; vision: world’s #1 AI content platform. Team from ETH, CMU, Meta, plus co-founder with ~10 years ops at US-listed VIPKid 1-on-1. Two consecutive multi-ten-million RMB rounds led by Lancang / BlueRun and Oufang Angel.
Q&A
1. Product and core scenarios?
(a) Wawa Writing (蛙蛙写作) — AI novel assistant across genres/styles with creator-friendly protection. (b) Pro writing: personalized generation across massive themes at scale. (c) Enterprise custom: private LLMs, AI KB bots, digital humans — tourism, film/TV, games, e-commerce, healthcare.
2. R&D / GTM; biggest 2023 challenge and win?
Weaver — creative-writing LLM family (Mini/Base/Pro/Ultra) with novel instruction synthesis; routing Agent balances quality vs cost. Ultra beats GPT-4 on writing scenes in their evals; first LLM natively supporting RAG + function calling (per CN). API open; paired with Wawa Writing + AIGC solutions platform. Challenge: building full LLM pipeline + model/Agent/KB stack from zero — now among few China full-stack firms; hard but worth it.
3. GenAI opportunities vs prior waves?
Waves follow shifts in key societal factors. Internet shifted connection/reach (social, ride-hail, delivery, ads). AI shifts productivity — displacing higher cognitive labor (new-media ops, CS, web novelists, illustrators, designers). Later: transport (AV), entertainment (AI companions, interactive shorts). Credo: develop AI to make life better — serve humans (productivity equity via writing assistants; entertainment equity via immersive virtual humans). Remember: market upside always comes from where key factors change (cites Andrew Ng).
4. China landing challenges?
Regulation and copyright — but opportunity still outweighs challenge.
5. Key product / GTM factors?
R&D: LLM tech moat, data moat, tech–market fit. GTM: not fundamentally different from non-AI products.
6. Competing with giants?
Stay vertical, stay focused.
7. Are LLMs still too weak?
Depends on layer. Base-model firms should chase AGI frontiers (2023: instruction following, Agents, RAG, multimodal — 2024 will bring more). Full-stack firms chase “almost reachable” scenes with data moats + TMF — e.g. ship open Autogen-style Agent frameworks while commercializing Copilot (AI draft, human edit).
8. Human–machine collaboration?
Like AV: today Copilot; later Agent (AI plans/tools/does most work; humans set goals & supervise). Automating enterprise flows is a megatrend — ignore it and become passive.
9. 2024 tech to watch?
Agents L3→L4; memory-augmented LLMs; multimodal progress; outside AI — AR/VR.
1 CN image; attach on ingest. Signup CTA in CN omitted.
— END —