Connect your Claude, Codex or OpenClaw for full transcripts on millions of podcasts. Two week free trial, cancel anytime.
Invest Like the Best (Patrick O'Shaughnessy) · Money & Markets · October 2026
Patrick O'Shaughnessy asked what she thinks is overhyped, and her answer was software companies custom-training their own AI models. She says the real pitch is "you're a real AI company now", and doubts that customers actually want to be handed a model of their own.
Custom training models for the marketing value.
Be able to say we post-trained a model on our stuff. Explain why.
Go talk to the inference providers and sometimes give them, depending on the provider, give them true CRM. They will say, like, look, the reason to post-train a model is cost. Some businesses have that need, but there is not a quality reason to do it. There's kind of a marketing reason to do it right now in this environment. And I think there's a lot of post-training for marketing reasons for investors and for customers, mostly for investors, actually.
If one of those people were pitching the other side of this, the value of post-training, what is the pitch to you? One of these companies comes to you and is like, we'll do XYZ for you. What are they saying and what do they say the benefits are?
You are putting a stake in the ground of being a real AI company. Customers will notice. I think that's the primary one. I do think there's a secondary one in some use cases of like a form factor thing. And I think especially with large enterprises, you want to deploy this piece of software wall to…
I was one of these people and I came into Vanta, what does post-training literally mean? What would it mean in your context? What are you doing? What are the steps that you would go through?