George Sivulka of Hebbia says ChatGPT handles single tasks well but breaks down on complex, multi-step financial and legal work trained only on public data.
Hebbia's grid interface, where every document is a row and every prompt a column, is the biggest AI interface change since ChatGPT, according to Sivulka.
Sivulka predicts a massive financial market correction when AGI arrives, and says private investing will change more than public markets.
Why Hebbia thinks chatbots cannot do finance and law work
Hebbia is betting that finance and law need more than a chat window. Their work depends on private documents and custom process across many steps, so Hebbia built an interface that runs agents over every file and shows its work. The payoff is hours saved, and it arrives only once teams change their habits.
Takeaways
Chat models fail at complex finance work. ChatGPT handles single tasks well but breaks down on multi-step work. It trains on public data, while finance and law run on offline, unstructured information and custom process. Sivulka says it would hand every VC the same investment memo.
Hebbia's grid shows the AI's work. Each document is a row and each column a prompt or agentic action, so users work alongside the AI. Sivulka calls retrieval the main limit for AI in finance and law. An AGI would rather orchestrate sub agents than process 100,000 documents at once.
Time savings are large and measurable. A buy-side data room holds 40,000 to 100,000 files. Sivulka claims 20 to 30 hours saved per deal and analysts 80 times faster. Screening covers 137% more opportunities. Templates, now near 2,000, include a screening one that measures time saved.
Adoption is a people problem. Sivulka calls it sociology and change management, not technology. Early-career analysts are becoming Hebbia native, and the best deployments pair early adopters with aligned senior and junior staff. Boards now ask for P&L impact after the experiments of 2023 and 2024.
Software should empower analysts, not replace them. Citing Steve Jobs on bicycles for the mind, Sivulka says Hebbia aims to empower people. He calls the feeling of discovery the best part of investing and wants AI to remove the tedious reading that drives smart people away.
Per-seat pricing is meant to drive usage. Hebbia charges per seat because consumption pricing is too early and discourages adoption. Sivulka calls Salesforce a faster horse and contrasts it with Hebbia's continuous redesign, invoking Henry Ford.
Scaling laws hold and DeepSeek is overhyped. Sivulka calls training and inference scaling laws mathematical properties that hold with enough data, and expects GPT-5 to beat GPT-4 significantly. He says DeepSeek shows China can make American inventions more efficient, not push the frontier.
What Sivulka expects next
A massive correction in financial markets arrives with AGI. His Turing test is whether AGI can beat human investors.
Private investing will change more than public investing. Hebbia aims to be the Bloomberg terminal for private companies, while PitchBook and Crunchbase mainly serve sourcing.
Tap a name to go there · tap a line to see why they are linked
On this episode 2
Show 1
Mega funds with deal records back to 1975 could benefit, though he thinks pre-LLM deal history may matter less.
Over 50% of global GDP will come from AI agents within a decade, as new value rather than replacement.