The Innovation Civilization Podcast
00:02:05 · George Sivulka
[00:02:05] S1: One of my favorite questions to answer, and I think the the best way to answer it would be with a story. I started Hebia, I think it was around four years ago now, and it was out of my Stanford PhD, where I was really obsessed with very big questions. And in the PhD, thought that the the best question to ask myself is what's the most important technology that will ever be invented? And over the course of my studies, I convinced myself it would be meta learning, which is... Was the idea of teaching a machine to learn how to learn. So it's this topic du jour in AIML at the time. In June 2020, while I was working on trying to build the world's first metal... Meta learner, OpenAI released GPT three. And the title of the paper where GPT three was released was actually large language models are meta learners. And I remember sitting back in my chair and thinking the most important technology of all time, this idea of of creating an artificial brain, a machine that could learn how to learn and could generalize to any task, it was gonna happen. And I I I called the chat GPT thing early, But I thought to myself, it's a fundamentally important technology, but technology is not a product. Yep. And so while we have this technological revolution, and it might be one of the most important ones of all time, someone will need to build a product to capture that. In the theme of innovation civilization, you could go all the way back to... There have been seven major technological revolutions in the history of mankind. It's been like fire. There's been the wheel. Could even go... There's been the invention of the capital markets. More recently, the adventure of the invent... Invention of computing. And I think that artificial intelligence or or really meta learning, which is what we're really talking about here, because AI has been around for a while, is the seventh most important technological revolution. And none of those technological revolutions were important without a product. Fire wasn't important without the torch. The wheel wasn't really that important without the chariot. Even computing. Nerds used computers up until Microsoft invented Excel, and Apple invented the graphical user interface. And those are products that capture the power of the technology. Yeah.
00:04:11 · George Sivulka, Waheed Nabeel
And Hebiah is making this bet that if AI is the seventh most important technological of all time, and maybe one of our final ones, what's the product that will capture that? It's not gonna be a chatbot. And that's what we try to do. [00:04:23] S0: That's quite interesting, actually. The fact that, at least for me and a lot of other people who were not close to UI, just like a lay person, believe the ChatGPT moment or the chatbot moment was where it just came to the forefront that, oh, shit. This is so real. AI is real, and it can learn and so on and so forth. I was hearing one of your other kind of previous podcasts. You were saying that you actually don't like ChatGPT or the the chatbot product, really. So do you see that this is not the kind of feature or user interface of AI?
00:04:23 · Waheed Nabeel, George Sivulka
[00:04:23] S0: That's quite interesting, actually. The fact that, at least for me and a lot of other people who were not close to UI, just like a lay person, believe the ChatGPT moment or the chatbot moment was where it just came to the forefront that, oh, shit. This is so real. AI is real, and it can learn and so on and so forth. I was hearing one of your other kind of previous podcasts. You were saying that you actually don't like ChatGPT or the the chatbot product, really. So do you see that this is not the kind of feature or user interface of AI? [00:04:55] S1: It's a very good point, which is that, yeah, chat seems to have captured this like zeitgeist for everyone, but is it the final form? And I I genuinely believe that chat is maybe the first iteration of the technology, but it's definitely not the last. It's very similar to a single threaded calculator, like a t I 84. So you could show someone a t I 84 in the nineteen hundreds, and they'd have a heart attack. But it didn't really realize the promise of computing until people invented the graphical user interface, the GUI, and then Excel. And similarly, I think that chat is like a single threaded. It shows people the power of meta learners, things that can generalize to any task, that could write a poem in any style. But when you wanna use it for serious work, there's a much richer, deeper, more process oriented interface that is needed, and that's what we're trying to build at Hebia.
00:04:55 · George Sivulka, Waheed Nabeel
[00:04:55] S1: It's a very good point, which is that, yeah, chat seems to have captured this like zeitgeist for everyone, but is it the final form? And I I genuinely believe that chat is maybe the first iteration of the technology, but it's definitely not the last. It's very similar to a single threaded calculator, like a t I 84. So you could show someone a t I 84 in the nineteen hundreds, and they'd have a heart attack. But it didn't really realize the promise of computing until people invented the graphical user interface, the GUI, and then Excel. And similarly, I think that chat is like a single threaded. It shows people the power of meta learners, things that can generalize to any task, that could write a poem in any style. But when you wanna use it for serious work, there's a much richer, deeper, more process oriented interface that is needed, and that's what we're trying to build at Hebia. [00:05:44] S0: So if I understand correctly, and I was just seeing some of your demos, I believe you guys went for the Excel spreadsheet style, right, where if a person wants to use Hebia, they create a row and dictate what they want in the columns. And they... The Hebia AI just sucks up all the data, does the thinking, and just gives the answers in the columns. Right? Is that in a nutshell what your user interface is currently and what your product is?
00:05:44 · Waheed Nabeel, George Sivulka
[00:05:44] S0: So if I understand correctly, and I was just seeing some of your demos, I believe you guys went for the Excel spreadsheet style, right, where if a person wants to use Hebia, they create a row and dictate what they want in the columns. And they... The Hebia AI just sucks up all the data, does the thinking, and just gives the answers in the columns. Right? Is that in a nutshell what your user interface is currently and what your product is? [00:06:10] S1: That's one of the easiest ways to think about it. But even more than just being a spreadsheet, we're thinking of different ways that you could orchestrate many agents. So a spreadsheet's one way where every cell could be an agent, and the cell represents the output of running an agent over some data. And that's our first iteration of our product. But there's also ways to use grids and other very human interfaces to portray the activity of many agents, to portray agents of agents, and to allow humans in an understandable way to build pipelines of data and pipelines of data processing. [00:06:44] S0: Yeah. That makes sense, really. And just for our listeners, can you tell us, like, what are you using heavier for? So you mentioned about meta learning that was important, and you mentioned about a product around meta learning. Right? And you mentioned about, obviously, the grid style approach, which is, I guess, like a first easiest way to think about it. But what problem are you trying to solve here that's not resolved with any other LLMs or whatever's out there? Yeah.
00:06:44 · Waheed Nabeel, George Sivulka
[00:06:44] S0: Yeah. That makes sense, really. And just for our listeners, can you tell us, like, what are you using heavier for? So you mentioned about meta learning that was important, and you mentioned about a product around meta learning. Right? And you mentioned about, obviously, the grid style approach, which is, I guess, like a first easiest way to think about it. But what problem are you trying to solve here that's not resolved with any other LLMs or whatever's out there? Yeah. [00:07:07] S1: Yeah. So we're just trying to figure out the ways to orchestrate agents for complex tasks. If you ask Chatuchipati a question, it'll give you one response. Or deep research might give you like a a longer response that takes a little bit more work upfront. [00:07:21] S0: Okay.