Episode Transcript
[00:00:00] Arshavir Blackwell: I've heard talk of ChatGPT models, very sophisticated ones, potentially being priced at a thousand or $2,000 a month. If you look at my one month, my Claude bill was almost $1,000. It was huge. The price that you're paying for those tokens is still heavily subsidized.
[00:00:20] Announcer: This is real AI use cases, business owners, roundtables.
[00:00:24] Elizabeth Gearhart: Yeah. There's one other question I had for you, Arshavir. You have your voice craft and what your information says is it's looking beyond the impressive output of AI to understand what's actually happening inside the model.
[00:00:36] Richard Gearhart: See, we should be using this and maybe answering with ChatGPT, you know, giving ChatGPT questions with ChatGPT answers. Maybe that's the solution. I don't know.
[00:00:48] Elizabeth Gearhart: How does your voice craft know what's going on inside the model?
[00:00:51] Arshavir Blackwell: Well, it's a little different from that. So the frontier models, what you have to understand is that the whole way that these systems are powered is they have kind of electronic guts, so to speak, and that those are in the series of literally billions and billions of numbers that we call the weights of the system.
The frontier models that are out there somewhere on a server and not confidential.
Those weights do not change as a result of, of what you put into them. So if you put in a prompt, you're always going to get very similar kinds of responses with respect to a particular input.
And there's really no way around that. Now, the prompt, you can do a little bit of prompt engineering and putting some stuff into the prompt that steers it into the direction that you want. But very often if you ask it to use a particular voice or a particular tone or a particular way of expressing itself, you will find that it doesn't come across as very authentic. It comes across as a computer trying to sound like it's talking about stuff with a southwestern drawl, let's say, with little fake cactuses in the background and not the real thing, in other words. And local models are able to be trained so you can actually change the values of those weights because you have control over the model, which you don't have. With front end models. We have a server that we use that's not ChatGPT or Claude, but our own server. And you can also just run it on your own computer if it's a fast enough computer.
And so what our system does is it takes lots of examples of your own writing, reads through them, trains on them in a way that frontier models just cannot do.
And then if you put in, hey, rewrite this, it will rewrite it in your voice because it's been trained on your voice. And that's the big difference. Now, there are other advantages to using it. One is you're not sending out confidential information to some server, ChatGPT, Claude, or, you know, any of those guys.
So you don't have to worry about concerns like confidentiality, hipaa, your marketing materials being leaked, all of that. The other thing is local models. You don't pay for the tokens because it's your local model. These huge token bills that people get go away because they're just using a local model and you can just keep hitting it as much as you want and all you pay for is the electricity.
So that's the idea of your voice craft, essentially. In a nutshell, it's a local model trained on your voice that provides you with documents and writing that sound like you wrote them.
[00:03:56] Elizabeth Gearhart: How much does that cost?
[00:03:58] Arshavir Blackwell: We're still working on pricing models now, but it's not particularly expensive. It's going to probably be around 30 or $40 a month for unlimited use or in beta right now. But it's not, you know, expensive by any means. If you look at like my one month, my Claude bill was almost a thousand dollars. It was huge.
And I use it for not just my own personal stuff, but I use it to help me with coding. It has agents as well. So I assign agents tasks like, hey, post this every day at 3pm Here's a whole bunch of different notes I want you to post. Go do that.
It will do that for you. But it does. If you rely on it as much as I do, it does start to get very expensive.
[00:04:42] Elizabeth Gearhart: Yeah, I had heard this story where someone in a corporate job blew through the whole cloud budget in the first quarter.
I
[00:04:50] Arshavir Blackwell: can well believe that because it's not cheap. The thing people don't realize is that these big AI companies, as great as their value are, they are really blowing through money. So the price that you're paying for those tokens is still heavily subsidized by
[00:05:09] Elizabeth Gearhart: investors. Right.
[00:05:10] Arshavir Blackwell: By those investors. And that's one reason why, I think, to go back to our first topic of hype and is this going to kill everybody? They really need to get people investing in what they're doing in order to continue on providing this service at the current rate that they do, at the current pricing model that they do. I've heard talk of ChatGPT models, very sophisticated ones, potentially being priced at a thousand or two thousand dollars a month. So if you look at how much Power and how much cooling these systems take to generate their responses, it really does get very expensive. I think that's going to go down and that's going to change. But that's definitely a factor right now in the dynamics of how they do the pricing and how they figure out exactly what your bill is going to be at the end of the month.
[00:06:08] Elizabeth Gearhart: Why isn't AI solving the data center's problem? Why can't it find a recirculating coolant so it doesn't need to be on a river? Why can't it develop superconductors so it can send and receive energy without loss? And why can't it take the heat and recycle it, dump it into some sort of battery system? Is AI working on that, do you know?
[00:06:29] Arshavir Blackwell: I don't know. Personally, I suspect that there are people that are working on all of those problems, and I'm sure they're using AI to do it.
So they're not easy problems to solve. AI is very powerful, but it's not a magic bullet. But I suspect that's actually a very good example of how AI can help AI.
The main problem with data center. And by the way, you know, data centers are not new. When you use the Internet, you're using a data center idea that all of a sudden they're building these data centers that never existed before. That's not true. What people are worried about is more energy usage, more heat being pumped into the atmosphere.
And the thing is that they use water, usually to cool them. And the problem is that there's always going to be some evaporation. You just can't get around that.
There have been plans.
One plan has been to put data centers in the Arctic, where it's already very cold, but you're still pumping heat into the atmosphere. The only other solution I've seen that I think is really it's not going to be happening tomorrow, but it's workable, is to put them either in outer space or the moon and to have nuclear generators on the moon or solar, if it's an orbital platform.
And then you don't have to worry about heat dissipation or any of those problems because it's out and outer space.
And what you do then is you have data centers that you just have these satellite links with.
[00:08:05] Elizabeth Gearhart: I don't understand how the heat would dissipate in space either, because there's no air.
[00:08:10] Arshavir Blackwell: I'm not a space scientist, so I can't tell you the exact details, but I know that they have ways of Dissipating the heat in space. They probably have fins radiating the heat outwards or something like that.
But the other thing that I think may help us with this problem is you have to remember that right now that data centers and LLMs use a huge amount of energy compared to the human brain, orders of magnitude more. The brain runs on something like 60 watts, but it has to basically run on just the energy that you get from eating and drinking.
And the point of that is simply that it there must be something solutions to what we're trying to do.
But the fact that the human brain does what it does so much better than LLMs so well is suggestive that this energy output problem is something that in some, at some point or another should be able to be conquered.
[00:09:21] Richard Gearhart: So I was just wondering where you see artificial intelligence going in the future and, you know, what new changes do you think are on the horizon, say within the next six months to a year?
[00:09:35] Arshavir Blackwell: Yeah, I think that technological progress is not something that's linear. It goes for a while and then there are these sudden jumps. And that's what we've seen with LLMs, where neural networks kind of were, in a sense, moribund for a while. And then a paper came out in 2017, and then ChatGPT came out in late 2021, 2022, and then all of a sudden there was this huge spike of activity in how AI was used in the business realm.
And things since that point have continued to go upward and improve significantly.
But I don't really know if I could say that there's been a huge jump at the same level that we had in 2022 when this first came out. If you look at what AI can do now compared to just a few years ago, it is very impressive, but it's not the jump that it was when you went from no AI to suddenly having a relatively simple AI system in 2022.
So I don't think we're going to see amazing miracles. Let's look back to the web and the Internet. I mean, it was something that kind of started off and then there was a slow build, and then suddenly everybody was on the web and everyone was selling on the web.
And so that also went in fits and bursts and you're going to see the same thing. So I don't think you're going to see anything magical in the next six, six months or a year. I think it'll be roughly an evolution of what we're seeing now.
But it's very hard to predict, you know, three years, five years, 10 years down the road. Because what you have to remember is that we've now gotten to the point where these systems are designing themselves and we're beginning to create LLMs that no human being really understands how they were created, because they weren't created by a human being. They were created by another AI system.
Once that chain gets started, I think that may be where you see a sudden burst of innovation and a sudden burst of capability. But, you know, it's just so hard to predict. We were supposed to have flying cars and robots serving us in the 70s or something like that, and that never happened. Yet no one predicted the incredible rise of personal communications devices like the iPhone, or of how everyone would essentially have a supercomputer in their pocket. So the things that you think are going to happen very often don't. And then the things that do happen sometimes are just a huge surprise that nobody really predicted. Arthur C. Clarke, he was ingenious. He predicted all of this stuff, but he was the only one.
So I think that's why three years, five years out from now, things will be vastly different.
[00:12:40] Richard Gearhart: It might be fun to have an LLM, read all of Arthur Clark's books and do some predictions based on.
[00:12:47] Elizabeth Gearhart: And Heinlein too. There was one where the robots that started out, this man and woman were shopping and there was this $40,000 diamond necklace. And he's like, well, you can get that one. She's like, I don't want it. I have so much stuff already. The robots do everything for us and I'm sick of it. And
[00:13:03] Arshavir Blackwell: I know what that was. That was Frederick Cole and that was Midas World.
[00:13:09] Elizabeth Gearhart: Yeah,
[00:13:09] Arshavir Blackwell: it was a bunch of short stories along those lines that he wrote. Yeah, he put it into a book called Midas World. It was kind of a satire, but that was a very funny book.
And the. Well were basically the people that drove around in K cars and the poor people were the ones that had to use the really high end stuff because. Because they were forced to consume at maximal effort. That was very good.
When I read Arthur Clark, it's amazing to me, or when I see sometimes you can find him on YouTube things that he said in these interviews. It's like, wow, he really. He knew we would have phones in our pocket and that we would not be able to escape. He knew that all of this stuff. There's actually a book he wrote in the 70s and he describes what really is an iPhone in detail. It's a handheld computer that has buttons on it that you can do any one of a number of different functions and he describes it in great detail. And when you go back and read it, it's like this guy was talking about an iPhone. And that was in 1975 or 76.
So Arthur Clark, I think, was an unheld genius. I think he doesn't get enough credit.
[00:14:19] Richard Gearhart: Maybe. But you don't know Arthur Clark was right until
[00:14:23] Elizabeth Gearhart: 50
[00:14:23] Richard Gearhart: years later. I think somebody did a study of Nostradamus and said, well, you know, statistically somebody could predict everything just of pure coincidence. But I don't think that
[00:14:36] Arshavir Blackwell: was the case with Arthur
[00:14:37] Richard Gearhart: Clark.
[00:14:38] Arshavir Blackwell: I think Arthur had more on the ball.
He had more reason to make the predictions he was making than Nostradamus did.
[00:14:46] Elizabeth Gearhart: Arse. We're going to have to wind up. I thank you so much for your time. Before wind up. What does your T shirt say? I see it looks like there's an equation on there.
What does it say?
[00:14:56] Arshavir Blackwell: Okay, have you ever watched the Twilight Zone?
[00:14:59] Elizabeth Gearhart: My God, I love that.
[00:15:00] Arshavir Blackwell: Okay, do you remember the one where the Kaniment come to Earth and they have a book that is
[00:15:08] Elizabeth Gearhart: serve humans
[00:15:09] Richard Gearhart: to serve
[00:15:10] Arshavir Blackwell: mango. People are like, oh, they're here to help us.
The book, this is what it says on the front of the book in the Twilight Zone. Serve man in Kanamith.
[00:15:20] Elizabeth Gearhart: Yeah, yeah. Serving man means how to cook.
[00:15:24] Arshavir Blackwell: Yeah. And then there's a great line at the end where she runs up to him as he's getting on the ship to visit the Kahnamid planet and she says it's a cookbook. And he gets pushed onto the. Of course, I've given it away for anyone who hasn't seen it, but it was 70 years ago, so
[00:15:40] Elizabeth Gearhart: I love that. Well, how can people find you if they want to talk to you or interview you or hire you or use your new business, your new platform?
[00:15:49] Arshavir Blackwell: Well, you can subscribe to my substack, which is the gateway to getting to me@subscribe. Inside the blackbox.
AI people can email me too at just my first name A R S H A v I r e.com and those are the two best ways to get a hold of me.
[00:16:12] Elizabeth Gearhart: Okay, well, thank you very much as you.
[00:16:14] Speaker E: You have been listening to Real AI Use Cases Business Owners Roundtable with hosts Elizabeth Gearhart, marketing expert and PhD researcher using AI every day. And Richard Gearhart, entrepreneur, business owner and intellectual property attorney specializing in innovation.
We hope you found this valuable. Join us again for more stories. Because the future of business is driven by AI.
