Unique Research · Insight / Unique Friend Q&A · January 12, 2024 · Guest: Huang Li, Founder & CEO, Jingxiang Technology (镜象科技)
Inventory type remains insight (Unique Friend series — not partner welcome).
Guest
Huang Li — Founder & CEO; Cambridge MPhil Psychology; ECNU external master’s advisor; 15 years big data & AI; former Tencent big-data director.
Jingxiang — AI psychological counseling: multimodal AI + LLMs for assessment, companionship, and therapy for students and broad depression/anxiety populations via AI digital humans. Multimodal psych assessment software holds NMPA medical-device certificate; ECNU-joint AI counselor EmoGPT — early China LLM psych app — live on Xuexi Qiangguo and Baidu.
Q&A
1. Product and core scenarios?
Aim: psych-counseling AGI via three vertical Agents — multimodal video-consult assessor; Emo-crowd AI counselor; AI doctor for mild depression/anxiety. Assessors in K-12/university mental-health screening; counselors for online heart-to-heart scenes.
2. R&D / GTM; biggest 2023 challenge and win?
Schools (hard need) going well — 100+ schools in Guangdong and beyond. Internet chat scene: big-platform talks, business-model polish. Challenge: tight customer budgets. Harvest: LLMs lifted UX — professionalism + flexibility now solvable by injecting team expertise into models.
3. GenAI opportunities vs prior waves?
Prior waves often redistributed resources (offline→online). GenAI is creativity/productivity-driven — creates new resources; freed labor moves elsewhere (medical Copilot → more patients; short-video makers → other services). Rebuild every human-needed industry. Barriers fall — a mini-game that needed 3–4 people in 2022 can be one person in 2023 → product Cambrian explosion.
4. China landing challenges?
Compute cost; weak consumer pay willingness (spend big, earn thin); lowered barriers → shell/mode copycats and fake-AI noise → bad money drives out good; some tech abused (voice clone fraud) risks blunt regulation.
5. Key product / GTM factors?
R&D: clear boundary vs base models — features a model update can erase aren’t moats. GTM: 2B needs professional credibility; 2C needs platform relations (traffic cut-off) and defense vs growth-hack teams.
6. Competing with giants?
Unique data, ops resources, or scene know-how giants can’t reach — be exclusive, familiar, or small.
7. Are LLMs still too weak?
Limits = prompts, models, data. Diagnose which: missing pro data (hard); model size/quality (compute/algo — soon fixed); prompts (engineering — solvable). Non-AI founders often misdiagnose.
8. Human–machine collaboration?
Steam trains coexisted with carriages; Tesla factories still ~25% human after years aiming for full auto. Question isn’t whether to transform but how — upskill staff to manage AI, build relationships, higher creativity.
9. 2024 tech to watch?
AI Agents; multimodal LLMs; embodied robots; MR (headbands/glasses) — fusions spawn new products/scenes.
1 CN image; attach on ingest. Signup CTA in CN omitted.
— END —