The Exchanges

Q robustness to error really matters here. So to give an example, AI today being applied to potential Drug candidates sounds like maybe a better option than actually prescribing a drug to a patient definitively. And I say for now, today, the, the models that we have, because this may change, but what applications stand out to you again today as near-term opportunities that maybe have some of this fault tolerance?

A I think what we're going to see is, you know, the initial tools that have this sort of level of fault tolerance being around customer engagement tools, for example, in fintech, where both banks and fintechs You know, can use these sort of models to basically automate what might be a CX agent in the past. But today you see a sort of virtual assistants when you log into your bank. And I think there's a level of fault tolerance in there in that, you know, while it's obviously not great if the customer doesn't get what they want right away, but there are still is going to be CX agents in the back, right? I don't think these people are going away anytime soon. And so, you know, they might be able to be plugged in should there be an issue versus something that You know, it's just more sort of higher value, right? Like a automatic underwriting or automated accounting where things need to be a hundred percent correct. So for automated accounting, for example, if things are incorrect, there could be huge potential implications, uh, like IRS fines.

AI assessment note: “initial tools that have this sort of level of fault tolerance being around customer engagement tools”

Q Right. And if they're beneficiaries, I'm also curious to know if it's in the backend and maybe they're aware, maybe they're not. Do you think that there will be a requirement, for example, to disclose that something is being run through an LLM or how do you see regulation maybe coming into play in this arena?

A Yeah, no, I think it's a bit too early to tell today, but I'm sure there will be cases where You know, let's say in a world where loan decisioning or tenant things like tenant screening are being fully automated by LLMs and say there are consumer issues with that, or, you know, it turns out that the models that were being trained upon were using sort of discriminatory data sets. There could be huge implications from, you know, folks like the CFPB and other regulators who cracked down on these types of companies. But again, I think it's just something that founders are going to have to keep in mind, given it's no one is doing that just yet. Right. So it's, it's kind of warning if you, if you are going to use these things to be aware of the potential regulatory implications.

AI assessment note: “I think it's a bit too early to tell today, but I'm sure”

Q the process, people were not a fan of, even though in that example, I think actually humans were part of that process, but What else do you think founders should be thinking about here? We've kind of touched on regulation. We've touched on the transparency that might be required. Is there anything else that you think founders should be really paying attention to as they're thinking about integrating this technology?

A Yeah, for sure. I think, you know, on the product side, I think these product leaps will likely come from tuning existing large models with four specific sort of commercial use cases with specific data. However, I think the one thing that founders can forget is that no matter the sort of platform shift, whether it's been internet, cloud, mobile distribution is always key. And so even with this new sort of product leap where maybe, you know, you have consumers flocking to the product because you are Advertising say, hey, this automated tax product is powered by LLMs and you're gaining a bunch of customers through that. In the event that this sort of product become or technology becomes table stakes. Well, then we're sort of back at square one, which is distribution is still key. We often say that it's almost more important than product, but hopefully that's not the case given sort of a lot of the last generation of financial services, fintech products were more financial engineering than software engineering. And so it'll be interesting to see if these new product leaps Can account for distribution modes that folks have had in the past.

AI assessment note: “whether it's been internet, cloud, mobile distribution is always key.”

Q how different circumstances would be the opposite, where you'd want to vouch and say, hey, look, look at this awesome new technology that we're using. Um, and maybe a, a silly example is just when you use ChatGPT, you know I'm using an LLM. You know that front and center. And so, how do you think about that? How transparent should founders be when they're thinking about implementing this technology?

A I think it's, you know, a bit of pontific pontification at this point because it's so early, but I think in some cases there will be interfaces for consumer to, you know, consumers to, for example, do their taxes via chat GPT like interface or for an earnings call to be automated for where an equity research analyst can basically use a chat GPT like interface to, you know, search based on specifically tuned enterprise data for that company. But I think on sort of like more sort of consumer fintech products or other products that we haven't even thought of, these might just be sort of backend products or backend platforms where the consumer is never really seeing the interface, if you will. They're just beneficiaries of that technology shift.

AI assessment note: “these might just be sort of backend products or backend platforms”