Paul Gu. Co-Founder, CEO Upstart. Episode 8.

Paul Gu is the co-founder and CEO of Upstart, the AI lending marketplace. He co-founded the company in 2012 alongside Dave Girouard and Anna Counselman and took over as CEO on May 1, 2026, with Girouard moving to executive chair. Before running the company he was its CTO, overseeing product, engineering, and machine learning, plus general management of the emerging auto and home businesses; he's been on Upstart's board since 2015. He pioneered Upstart's statistical models for predicting income and employment, came out of the quant finance world, and is a Thiel Fellow and a Yale dropout. In this episode we discuss: 00:00 — "The model's gonna suck": why you can't copy Upstart 02:02 — Three sheep, two breeds, one A/B test 07:56 — Upstart as a position, a first derivative, and a second derivative 08:13 — The core business: 50 million Americans the prime market won't serve 10:31 — Why a loan P&L looks nothing like a shoe P&L 12:03 — Not lending off your own balance sheet 13:03 — Risk capital deals and buying commitment from your funders 15:32 — "An operating business that feeds a credit hedge fund" 16:02 — Where did yield go? Treasuries, inflation, and J&J bonds 18:19 — Where should a normal person park cash now? 21:20 — Inflating away the debt vs. the hard choices nobody will make 24:10 — Can AI get us to 5% real GDP growth? 26:06 — The distortions poisoning America: the GI Bill and housing 28:38 — Housing as a status good, and the real supply constraint 29:58 — Cartoon math: $100 of loans, $5 of contribution, a 20% fund 33:08 — Can an LP actually get these returns? Leverage and the Goldman piece 34:31 — Contribution margin vs. net income at a 35% CAGR 37:07 — Auto and HELOC: from -150% margins to breakeven 41:33 — The real moat: Jane Street techniques pointed at consumer credit 43:41 — The credit trifecta — growth, performance, and profit at once 45:30 — The five-year gauntlet in full: bad loans and angry regulators 48:09 — Building the foundation model for credit 49:56 — The four things you have to prove to launch a new product 50:23 — "You pitched me on being an indentured servant" 51:25 — Buying Prodigy, and why rewriting dealership workflows failed 56:08 — Phase two was also wrong: the lowest rate isn't enough 58:03 — The golf trip problem, and paying dealers to pick you 01:01:04 — Why Upstart stays away from new cars 01:06:03 — "No one wants to fund a new product. It's like nuclear waste" 01:06:33 — R&D loans on the balance sheet: paying for training data 01:08:11 — Pulling five years of signal out of six months of payments 01:10:52 — "There are no heuristics in our business" 01:11:58 — Rewriting XGBoost around net present value 01:14:04 — Second derivative: focus, hurdle rates, and one strategy per bet 01:16:51 — Killing auto refi, the product they were best at 01:22:03 — "Waiting is the death of execution" 01:23:19 — The dyssynergy of carrying four unprofitable businesses 01:24:42 — 2040: relevant to economic history, and unambiguously pro-human 01:28:46 — Does the US government need an Upstart? 01:29:21 — Jack in the Box, and "why don't you get a horse?"

GPT-4正在为你翻译摘要中......

翻译全文