Highlights
-
2024-11-04 20:25 For too long, sustainability has been someone else’s problem. We think it should be our problem and we should lean in. So we’re creating specifically best practices frameworks. We established a standard and we’ve got a version of our standard published that you can consume. We’re pursuing creating certifications in this space.
-
2024-11-04 20:22 So we’re having a lot of conversations about ethical AI. We’ve actually just released a white paper and are coming out with a Conversation about sustainable AI. Okay, so seeing as you brought it up, what are you calling sustainable AI? Yeah, so I say AI is we’re early days. People have seen it do magical, interesting things. I can go create deep fakes, I can create great content. It’s got some issues as well. So we’ve come out with essentially a framework. And the framework trying to be catchy, memorable is reflect, reframe and reimagine. And the idea is. So reflect is pretty simple.
-
2024-11-04 20:23 And by 2050 we think it’ll be double what it is today. And 2050 isn’t that far away. So the other really disappointing fact is less than 20% of that is formally recycled today.
-
2024-11-04 20:23 Technology is it like engage with a partner. We’ve got a partnership with a company called Human AI, great organization globally that will partner with companies and help solve this. Some of the E waste stream problem, recycling or more importantly reuse problem.
-
2024-11-04 20:22 So they’re doubling every three and a half, about 3.4 months. I mean, it’s just crazy curve. Even GTP3 used about 1.3 gigawatt hours to train. GTP4 is 10x larger than that. So GTP4 used the minimum power as a small city for about four months of a small city’s power just to train that model.
-
2024-11-04 20:23 There are a bunch of things researchers are doing to improve on it, but both because it’s good for the world and good for the economy, but also, frankly, it’s good for their bottom line because it costs literally billions of dollars to train these models. Hey, what if I could figure out a way to do it more efficiently? And so one of the things today is these models, on a very oversimplified level, are taking one word in a document, kind of correlating it mathematically to every other word in the document. They’re going to word number two and doing the same. Well, what if I knew three steps in which branch was going to be a dead end and I could prune that branch and not train that branch.
-
2024-11-04 20:24 So using techniques like that, you could save, literally you can be 80% more efficient of power consumption and get to about a 90 to 95% is a good model, and maybe that’s good enough. And certainly considering the money and the power savings, it’s certainly worth considering that people are looking at that. There are also solid state models where they’re training very differently that are much, much less energy.
-
2024-10-30 20:41 Obviously not everybody has that kind of lever, but it does remind you that whole never waste a good crisis kind of advice.
-
2024-11-04 20:24 AI is supposed to be so smart. And they looked at a bunch of data they gathered into their AI model and then trained it and did some work and ultimately they came back with an idea that they hadn’t thought of. And the idea was, well, what if instead of just trying to buy reduced power or buy green power, what if we could essentially live in the valleys of demand? And by that typical power grid there’s peaks of demand and then there’s valleys of usage. And the notable thing about the power grid certainly in most developed developing and developed countries is there’s no buffer capacity.
-
2024-11-04 20:24 Yeah, there’s a former Tesla, one of Tesla’s early battery scientists and engineers left Tesla and formed its own company. And what they’re trying to create is essentially economical battery platform. So there’s more kind of battery storage distributed around neighborhoods. And it’s using a very different kind of battery. It’s not the nickel, it’s not the lithium and kind of rare metal kind of stuff. It’s using literally rust, it’s using iron Oxide to create these much cheaper, easier to store kind of batteries that have longer life and are kind of environmentally more friendly, but may help solve some of this grid, this kind of grid of balance problem. Yeah. Iron air batteries, aren’t they? Yeah, heard about them. Very good.
-
2024-11-04 20:25 That’s a good question. It’s bit left field. You know, there were so many kind of celebrity spokes, people that I think, you know, on certain topics people, people get, aren’t, aren’t particularly amenable to that. If I look back Historically, what, what I would say is, you know, somebody politically, if I go back to like, like the U.S. president Roosevelt, who looked at a big problem, you know, the, the global recession, Depression, Great Depression, and didn’t try to solve it in a kind of haphazard way, he thought, really thought it through, said, hey, we’re going to create a social safety net. We’re also going to put people to work to do infrastructure projects. We’re going to pay them for that work. They’re going to have pride in ownership of what they’ve built. They didn’t get a handout. So he fucked about the problem in this sort of circular way. He looked kind of at the circular economy. And so how do I solve all the problems we’re having, not just the surface level of the problem. And that’s really what we’re trying to do, look at the holistic problem and provide holistic solutions, not just easy solutions, kind of off the top. Nice. Good one.