Decentralized AI

“My mission in life: in the next 12 months Akash will be the best platform to deliver AI workloads, like centralized, decentralized, [whatever]… because that AI is too important to be centralized.” — Greg Osuri, 00:46:35

“The major, major, major milestone I think the next 45 days is going to be Akash AI… It is going to be the best AI developer product and that’s the goal. It’s not about non-custodial nature, it’s not about crypto… AI developers are not going to use Akash because it’s decentralized; AI developers are going to use Akash because they get cheap GPUs and good developer experience.” — Greg Osuri, 00:43:52

“Web3 is incredible to attract supply because it has [an] amazing incentive model, amazing network effects, amazing distribution mechanism to attract supply — but terrible at delivering value… the web3 [is] the supply, the AI is the demand… Did you see the recent ChatGPT numbers? 1 million users within five days and 100 million users within 45 days — the growth of demand is incredible.” — Greg Osuri, 00:38:26

“In the future [access] is going to be from each other, because right now… it’s impossible to get them, not even from cloud providers, unless you know someone very higher up… I think if you think GPUs are going to be the power of the future, we need that to be decentralized, and that’s what we’re working on.” — Greg Osuri, 00:04:23

“Decentralized machine learning is extremely complicated… these machines, in order to learn, need to be very, very fast… it’s very hard to do decentralized machine learning, right? That’s why machine learning is very, very centralized.” — Greg Osuri, 01:02:16

“[It] has to be decentralized, has to be in the hands of people… The fuel for AI is GPUs. If you take away GPUs, [AI does] not exist, and whoever controls GPUs controls the world.” — Greg Osuri, 00:45:12

“Their community just celebrates stable diffusion and… render entering the AI era, where in reality Render can run one model; Akash can run about 15,000 models on Hugging Face — any model on Hugging Face you can run on Akash.” — Greg Osuri, 00:48:21

“We’re starting the incentivized testnet in a week or so — we have about 300 signups right now… a lot of them are doing machine learning, so about 150 new machine learning applications soon. So as we launch our mainnet I think we’re going to see an explosion in machine learning, because the value prop is very clear and demand is very clear.” — Greg Osuri, 00:51:55

“With AI, with the way we’re going… a permissionless, decentralized GPU network is going to be so critical to power this enormous demand that we’re getting with AI. A lot of times it’s just interacting with a chatbot… but you’re talking to an AI and that AI is using Akash in the background. You’re not going to know — you shouldn’t know. Most people don’t know Netflix runs on Amazon Web Services.” — Greg Osuri, 00:26:40

“Yes, it is possible to distribute the training across different GPUs. In fact there is a proposal on Akash discussions talking about how they plan to use about 24,000 A100 GPU hours to train across a distributed cluster… so yes, it is possible to train on geographically distributed clusters on Akash.” — Greg Osuri, 01:20:48

“One attack on Render, one state attack on Render, well, if OTA, the company behind Render, it disappears. The Render network disappears right now, right? If Akash, the company behind Akash’s Overclock Labs disappears, Akash will run, right? Overclock Labs operates one provider out of 50 other providers.” — Greg Osuri, 00:47:03

“Decentralized open systems, even though they move slower, they go further versus closed source systems move faster, but there’s a limit as to which they can scale, right? There is nothing more scalable than the worldwide web. Why? Because it’s decentralized… it’s actually going to be decentralized solutions. I have very high confidence in that.” — Greg Osuri, 00:26:39

“The race to AGI is on. The question is, who’s, is that going to be a centralized or a decentralized player? I want more equitable. I don’t want a single corporation. I want the common to own the AGI… I want to see decentralized machine learning take a central stage.” — Greg Osuri, 00:26:39

“Today I saw some stat… it said one percent of global GDP will be used for machine learning very soon… I want Akash to have a sizable market share of the one percent of GDP. That’s my metric of success for Akash.” — Greg Osuri, 01:35:03

“It’s funny that Akash seems like GPUs overnight, but if you look at our white paper, the first section that talks about motivation — we talk about Akash, it’s primarily built for machine learning. This is published in 2017, 2018.” — Greg Osuri, 00:24:21

“Their business model is to make money on you. You are what they’re selling… you are what is training ChatGPT. The question is, do you want to do that and live in a world [of] feudalism — that means very few people having a lot of control and a lot of power over freedom of choice and liberty?… Akash has to be successful for this future that we want.” — Greg Osuri, 01:38:36

“We saw a full foundation model proposal to run on Akash that consumes about 24,000 A100 [hours] — the first time ever a foundational model is being trained on a decentralized cloud. The notion today that foundational training is a lot better on centralized versus decentralized is going to be debunked.” — Greg Osuri, 00:28:57

“GPUs are the new oil, and whoever has GPUs essentially controls their destiny in terms of their AI. I’m a big believer in sovereignty when it comes to AI. I’m a big believer in decentralized compute networks where I should own the A100s or H100s that train my AI, and I want to control the cost.” — Greg Osuri, 00:36:08

“We essentially theorized this supercloud idea in 2018 in a white paper called Akash Network… we were the first ones to even introduce the idea what a functional supercloud would look like, and Akash I believe is the only supercloud as far as I know.” — Greg Osuri, 00:22:00

“If you’re telling me the gatekeepers of this AI — which is essentially the people that have GPUs — are someone that has their shareholders’ interest at their heart, versus the consumer or the society or humanity… [that] is troubling for me, a lot of us.” — Greg Osuri, 00:15:29

“We’re also seeing some of the decentralized networks now training on Akash, right? So like BitTensor and networks like that. So Akash, I mean, I predict Akash to be the platform of choice for decentralized machine learning to thrive.” — Greg Osuri, 00:22:10

“If you ask me, I think it’s going to take a while by the time you can see something practical in this area… open source AI is doing well and open source AI is not on-chain AI — Akash is great for open source AI. We’re at least like five to six years away from on-chain AI.” — Greg Osuri, 00:56:58