Eilon Reshef: preference

Transcript context

…Awesome. Okay, let's talk about AI for a bit. You guys were very early on AI, you were working on, basically AI was like, your product was built on machine learning back then is what it was called before. Because it was cool and everyone probably thought it was a waste of time, and no, it's never going to work. Now, everyone's building AI, building AI into their product. What have you learned about working with AI over the years that you think people maybe are not yet aware of, or that will likely cause them pain that you can help solve and avoid for them? Yeah, funny, when we launched Gong, we didn't use the term AI because people thought it was a bad thing, it makes wrong decisions, or they just thought it's an [inaudible 00:32:26] item, an acronym. When we founded Gong, I was in sabbatical, and I actually went to this deep learning course because I was bored, in all fairness, and after that course I ended up buying Nvidia stock, which I wish I had kept up until now. But I did send an email saying, hey, this is the next thing. So, we understood it's the next thing, of course we didn't know it's going to be LLM and GPT and other acronyms that evolved over the years, and probably now that we're talking end of 2024-ish, I think people should not go from one extreme, which is, hey, we need a bunch of data scientists for every small project, which was the case five years ago, or three years ago, to the other extreme, which is, hey, LLM is going to solve everything. Because LLMs don't solve everything, they have huge utility. We use LLMs all over the place, most companies that developed AI stuff use LLMs, it's a great thing, but at the same time, don't assume it does everything, you still need some of the core competencies of AI. So, you do want to have expertise, people who actually know what they're doing, and help guide us PMs around, is this something that can be built, or no? Because if you're going to spend many, many hours on asking an LLM to do, I don't know, in the case of Gong, for example, tell me what the good sales cycle looks like? LLMs don't do that, it's just like maybe something else does, but we have a field prediction model. LLMs cannot predict fields, it's like we're very, very specialized. So, I think you still need to have expertise, you still want to have some measurements. So, yes, version one, you can just go to an LLM and say create something, I don't know, whatever, but if you don't have measurements... Like in the old machine learning, whatever metrics you use, you're not going to advance, you're going to have V1 and you're going to have V2 and you have no way to know if you've made a progress. So, we pay a lot of attention to, we have people who are going to specialize in how you measure, we use ELO system, which is the one used chess as well, and we do have experts who can help us make the right decisions. You can make a very good progress without these, but I think there's a glass ceiling if you don't figure out how to create a more operational rigor around this whole AI thing. So, what I'm hearing is don't assume you can just outsource all your AI magic model building to the foundational model companies, you need to have your own AI expertise, ML expertise?…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.