Unique Research · Unique Partner · June 26, 2024
AIWaves Inc. (波形智能) is a Hangzhou AI company whose long-term mission is “realize AGI and make life better,” aiming to become a leading provider of accessible AI applications. On its self-developed infinite personalized content-generation model Weaver, it builds consumer product FrogWrite (蛙蛙写作) and offers vertical LLM customization for enterprises.
Jiang Yuchen, founder of AIWaves, trained in Zhejiang University’s Chu Kochen Honors College engineering-management program and earned a PhD in CS from ETH Zurich under Prof. Ryan Cotterell (controllable NLG, long-form generation, MT, coreference). He has published at ACL, ICML, EMNLP, NAACL; organized WMT Shared Tasks; and researched LLM training/inference at UCLA, Japan’s AIST, and MSRA.
Q&A
1. Core products and scenarios?
Two products: consumer one-stop creation tool FrogWrite, and enterprise DevTool AIWaves LLM/Agent Platform. FrogWrite centers on novel writing, with BOT tools for creator content, marketing, e-commerce ops, workplace docs, official writing, teaching, and homework. The platform targets enterprises that want to build AI apps — enterprise knowledge bases, vertical model training, and Agent-building DevTools.
2. Commercialization?
FrogWrite: membership subscriptions, plus à-la-carte features (e.g. book-splitting in story mode) and multimodal image/video (tweet videos, talking-head clips). Platform: annual enterprise subscriptions — clients include Kuka Home (顾家家居), covering bid generation, official docs, knowledge bases, AI coaching, and AI marketing.
3. Product and business-model innovation?
Personalization first: FrogWrite stores positive/negative feedback so outputs track the user’s style as they write more. A lively feedback community closes a win-win loop with constructive reviews on every release.
4. Staying ahead as foundation models iterate?
Technology foresight plus a long R&D plan. Three principles: (1) “Don’t stand in OpenAI’s way” — build tech that compounds with L0 Scaling Law (e.g. their infinite personalized generation / LPA) rather than getting overwritten when base LLMs get stronger; (2) play to team strengths — multimodal matters, but they watch and call multimodal APIs rather than rebuild the stack; (3) seek certainty amid uncertainty — invest early in compute, frameworks, and DevTools (SOP-based Agent framework, LLM factory) that every app needs, commercializable beyond their own C-end apps.
5. Control when apps depend on third-party “black-box” LLMs?
They self-train models, so product control is less of an issue. R&D couples tightly to scenarios: hallucination hurts Q&A but not creative writing, where users want style and one-person-one-face personalization — so they invest there. Scene-led, demand-pulled tech is the team’s edge.
6. 2024 trends worth watching?
CN source ends at the signup CTA for this question — no answer text in the archived article.
4 CN images; attach on ingest. CN inventory type=insight; EN typed partner (非凡伙伴 Unique Partner spotlight).
— END —