Unique Research · Insight / Unique Friend Q&A · January 16, 2024 · Guest: Wang Shouyan, Founder, DeepInsight (ex-Baidu big-data base platform founder)
Inventory type remains insight (Unique Friend series — not partner welcome).
Guest
Wang Shouyan — Founder, DeepInsight; founded Baidu’s big-data base platform. DeepInsight redefines commercial BI with AI — product DeepBI. True BI, in their view, means AI fully replacing humans for analysis and mining. Beyond LLMs for “learning,” they stress collaboration and thinking via multi-agents toward AGI. Ultimate BI goal: deep insight via deep thinking — LLM agents as infinite-depth thinkers; BI is the entry, not the ceiling.
Q&A
1. Product and core scenarios?
DeepBI for commercial data analysis and mining.
2. R&D / GTM; biggest 2023 challenge and win?
Core modules done and open-sourced on GitHub; more industry data tests for fit and performance; marketing-driven commercialization. Challenge: redefining a decades-old BI market requires being 10× better than classic BI — otherwise newcomers have no chance. Harvest: qualitative jump in AI base capability enables products classic BI never dared; AI-native development lets ≤10 core engineers ship what traditional IT needed dozens or hundreds.
3. GenAI opportunities vs prior waves?
AI-native IT rebuilds code-centric development — every industry’s software gets rewritten. Largest software-layer revolution since computers — and the biggest software opportunity.
4. China landing challenges?
Few base models near GPT-4. Today: GPT-4 fully, or API routing mixing GPT-4 with others. Complex scenes without GPT-4 are often unusable; no full domestic substitute yet.
5. Key product / GTM factors?
Must be 10× better than traditional software or users won’t try/migrate. Product capability is the key — GPT-4 itself markets cheaply because it is 10×.
6. Competing with giants?
ToC may face giant pressure (that’s how they became giants). ToB: many industries; giants can’t cover all and often show arrogance in service — strong product = many openings.
7. Are LLMs still too weak?
In practice GPT-4 is enough for their complex analytics; other bases usable if they hit GPT-4 within ~1 year. Hallucinations/drops fixable with engineering. They don’t expect LLMs to do deep collaborative thinking — strong knowledge I/O and logic suffice.
8. Human–machine collaboration?
Principle: product features are AI self-serve; engineers are coaches. New industry data → AI finds methods, sets flows, codes/debugs, analyzes, even writes PDF reports. Talent skews to thinking ability (train AI to think — need people smarter than AI). No classic PM/UI/FE/BE/QA split — AI took most of that work.
9. 2024 tech to watch?
Whether domestic closed/open models catch GPT-4 — widens commercialization choices.
1 CN image; attach on ingest. Signup CTA in CN omitted.
— END —