Ying Hua, cofounder of Implied and former Balyasny portfolio manager, argues general model companies will not eat the vertical finance market because they are unlikely to do the slow work of ingesting ever-changing market knowledge.
Ying Hua says Implied processes its own data, running live transcription with under one second of delay and a scraper across seven highway patrol websites that tracks accident duration for car insurers.
Ying Hua argues static dashboards fail because analysts tire of them, so AI should act like an intern that reads dashboards and flags what matters, while Implied keeps judgment with humans.
Why domain knowledge, not general models, will win AI investing
Ying Hua argues that AI investing tools win through ticker-level domain knowledge and carefully processed data, not through general models. Language models match patterns and repeat what they are fed, so they can gather and organize verifiable facts but should leave judgment to human analysts. That is why Implied built its own data pipelines and an AI-native spreadsheet rather than dashboards or Excel add-ins.
Takeaways
Public stocks are an especially hard domain for AI. Information shifts with a single news item, the market is tightly interconnected, and understanding must span qualitative and quantitative analysis. Ying Hua says this makes stocks harder than AlphaGo or poker, so generic models fall short without deep domain knowledge.
Vertical finance tools will survive general model companies. Ying argues horizontal labs will not eat the finance market over the longer term, because they are unlikely to do the slow work of ingesting ever-changing market knowledge. Models are like new graduates needing extensive training on a team.
Models predict patterns and should not own judgment. Ying calls language models backward looking next-token predictors, not true reasoners. They repeat management claims and misread metrics, and guidance is missed about 30 percent of the time. Implied sticks to verifiable facts and leaves judgment to humans. Khe Hy countered that sheer pattern volume may amount to understanding.
Institutional-grade data must be processed, not scraped from the web. Free sources like SEC filings and transcripts are only a starting point. Implied runs its own live transcription with under one second of delay, keeps filler words for sentiment, and uses agents with humans cleaning data. Brett Caughran says the open web is not institutional grade.
Analysts cannot easily replace data engineering with Claude Code. Agents make scraping easier, but non-coding analysts still struggle with scheduling, storage, website changes and checking logic. Ying says a non-technical person can hardly verify AI-generated Python. Her own car insurer scraper covered seven highway patrol sites.
Static dashboards and Excel add-ins are the wrong design. Analysts abandon chart books around the sixth one, and dashboards will follow. Ying wants AI to act as an intern who flags what matters. Implied built a cloud AI-native spreadsheet that auto-updates on earnings, so the agent only remembers formulas.
Always-on assistants will let one person cover more companies. Ying predicts the next step is an always-on assistant that learns from feedback rather than a single chatbot. One person with AI can cover far more companies in more depth, and one client already relies on Implied's chat instead of Excel for some names.
Tap a name to go there · tap a line to see why they are linked
On this episode 3
Show 1
Where Ying Hua sees AI investing going
An always-on assistant that learns from analyst feedback replaces the single chatbot.
One person with AI covers far more companies and does deeper analysis.
Spreadsheets that update automatically when companies report reduce the need to open Excel.
Brett Caughran plans to have Ying back in a few months to check whether these predictions came true.