Open-Source AI

“Akash is fully open source software. The company Overclock Labs makes no money on Akash — we built the software, we gave it out for free… the best marketing always should come from our community.” — 00:37:07

“You can run your own Stable Diffusion — I’m very bullish, because it’s the first open source AI model. OpenAI and the Google stuff is closed source. That means anyone can run Stable Diffusion on any computer that has sufficient GPUs, or Akash.” — 00:38:31

“It’s very important that we get this open source. If you deem AI as one of the most revolutionary technologies that’s going to benefit humanity, it’s very important that we get it right — it’s very important that people own it and not owned by companies… Think about surveillance capitalism now, from Facebook and Google, to the world — think about surveillance capitalism when machines take over.” — 00:38:13

“We went radically open… we open sourced the process itself… everything we do except for personnel issues is open… right now there are about 229 active contributors to Akash, [of] which 20 of them belong to Overclock Labs — 90% of contributors for Akash Network are outside the company that created it… Overclock can disappear today and Akash will still operate.” — 00:42:43

“AI is undoubtedly going to become the most important technological shift that has ever happened to humanity… you have OpenAI, which is not really open source at all, just the name is OpenAI… sucking the air out of the open source AI movement. So I think for open source AI to exist, we need to decentralize the access to that open source AI.” — 00:10:46

“No matter the advancements, the models are not going to be translated into actual results, in actual usable things people are building… I think open source AI is going to be the future.” — 00:20:10

“If you’re a company you can move fast, right, but… you can’t move as far as you can with the community. So I think open source will eventually win out… it’s only a matter of time. And also products haven’t really caught up to how fast these models are moving… by the time products really catch up to GPT-4, I think the open source [will catch up].” — 00:18:42

“Regardless of what the founders intend the company to be, the structure is created in a way that people are only incentivized to increase shareholder value… With closed source software, we can go fast. But with open source software, we can go further. And that’s the core difference.” — 00:19:47

“If you want to create [something] hyper-scalable, that has network effects, that’s purely open source — decentralization is the only way. I can challenge anybody to challenge my assumption… Crypto adds a network aspect to open source that never existed before… Open source can go farther where closed source can go faster, but not as far as open source can.” — 00:44:46

“It’s not how advanced or how new a model that determines the quality, it’s really how you train it. So SLMs are taking a center stage now… you can deploy an SLM and train an SLM on Akash.” — 00:28:02

“There was a board takeover. Five people that nobody heard of before were able to control the most powerful AI in the world. Think about the implications of that… imagine if they had access to AGI… There is inherent fragility in these old systems that need to be changed and the only way to do so is through open source. Closed systems can go fast, open systems can go far.” — 00:35:57

“Crypto is actually a way to solve open source sustainability. Even companies like Docker couldn’t exist because they were following the old model — the old model doesn’t work anymore. Docker ran out of business, they sold… for peanuts, and Kubernetes is having a crisis with their contributors… because there’s no way for people to work on something full-time and get paid for it.” — 00:26:41

“I’m a big believer in open source software. Why? Because with closed source you can go fast, but with open you can go far. Open systems building upon each other is how we can go fast… Akash is essentially another command, part of the Bittensor operating system.” — 00:18:10

“With Llama 3, open source has gotten as powerful as closed source now, so there’s nothing that has to stop [us] now.” — 01:00:47

“We have an enormous community, about 500 open source developers now contributing to Akash. We want to make that 5,000 by the time we have the next Accelerate.” — 06:38:39

“If that can happen before the Metas of the world stop releasing their frontier models open source, then we have a very good future ahead of us. If that doesn’t happen before those companies stop releasing the open source models, then we have a possibly very dystopian future in front of us.” — 00:18:47

“So AI will be open source and decentralized. It has to be open source and decentralized. It’s too important for it to be closed source and be controlled by a few elites.” — 01:04:56

“The foundation of our technology has to be open source. What you build on top of that, you know, can be closed source, can be open. It’s up to you.” — 01:14:14

“Google would not exist today if not for the first server they built in Stanford using Legos and Linux… So without open source, Google would not exist. So imagine how many Googles are we denying today with closed AI.” — 01:17:07

“There was a publication… they talked about how Google basically said they have no moat with open source AI… When that happened, open source AI was maybe about a year behind close source AI. Now it’s three months behind… Some models are way better than close source models. Why? Because when you open source, you get more talent looking at the model.” — 01:08:33

“Open source always wins out in the end. Right? It may take longer, but it will always win out. That’s why I have this saying that, hey, close source, you can go fast. With open source, you can go far.” — 01:09:58

“That’s why one of the reasons I was at the Hill fighting to keep AI open source — because when AI is open source, we can at least point out that this was used against a victim, and we can easily trace it back. If it’s closed, we don’t know what exactly is going to happen.” — 00:42:17

“I agree with the small models taking over, because we see the data in Akash as well… Have you noticed ChatGPT forgot how to code?… It’s not really how big a model gets that can evolve to AGI. I think it’s going to be the smaller models that are way better at coding or performing individual tasks.” — 00:26:28

“I think he’s just there to make money. I know this for sure because we do a lot of policy work in DC, and every time I go there, Sam Altman and the gang is trying to shut down open source and decentralized AI actively through lobbying, and they try to pass laws in the US to make open-source AI illegal, because they want to keep AI to themselves.” — 00:07:14

“Most of the legislators and the staffers we speak to, all of them agree that AI being open and decentralized is better for America and the rest of the world, and unless there’s concrete damage that’s going to happen, they’re not going to write rules.” — 00:07:57

“Docker, as a company — a tool which everybody uses… couldn’t survive; it raised a billion-dollar valuation, it took the web-two style of SaaS businesses, but that doesn’t play out well for open source software. How do you have a sustainable model for open source software? Well, decentralization… We have about 500 contributors, and that’s not possible using traditional centralized systems.” — 00:10:02

“There’s this famous article from Google that leaked that said Google has no moat in models, because models are getting open source… Just like how Linux took over the internet — now 96% of web servers, the top 1 million websites, are run on open source Linux — similarly we’re going to see that for AI as well. Initially it’s going to be centralized companies that have an edge, because closed source tends to go fast whereas open source tends to go far… but eventually it’s going to be open source and user-owned future.” — 00:03:41

“Moderation is done by the providers and not by the protocol, and the protocol should remain free of moderation, free of censorship, and permissionless — that’s the only way to go far. Permission models and closed models can go fast in the near term, but open and permissionless models can go far, in places where traditional software cannot. One of the examples I give: Akash is being used by DoD, Department of Defense, because of its open source nature and permissionless nature — of course a fork of it, not the actual network… and we have zero salespeople.” — 00:14:38

“A few companies write models, like OpenAIs of the world, Anthropics of the world, but really there are 40 million developers that could essentially write models. Models are not a lot of code — 600 lines or a thousand lines… so decentralizing that intelligence to the people using blockchain is a phenomenal way.” — 00:02:14

“A few companies write models, like OpenAIs of the world, Anthropics of the world, but really there are 40 million developers that could essentially write models. Models are not a lot of code — 600 lines or a thousand lines… so decentralizing that intelligence to the people using blockchain is a phenomenal way.” — 00:02:14

“It’s very obvious OpenAIs and Microsofts and Amazons of the world are fighting to keep AI closed, because open-source AI fundamentally threatens their existence. Because open-source AI is the biggest enemy of OpenAI, which has nothing to do with open-source AI… OpenAI is open only in its name.” — 00:15:01

“The general consensus we spoke, for most sides of the party, both Democrats and Republicans, is that AI, when it’s open, when it’s more accessible, benefits most Americans and the world as a result, versus being guarded by the few… unless there is concrete damage in terms of open-source AI that we can look at, we’re not going to get regulated.” — 00:15:43

“Yes, you can create companies as a closed source system, but in order to create communities and protocols you need to be open sourced, and that is the foundation of decentralization.” — 00:19:30

“The open source AI is maybe two to three months away from catching up to closed source AI. It’s very good — in some cases, like Llama 3.1, which is a 405 billion parameter model, exceeds the quality of closed source AI.” — 00:08:00

“The main thing for me for decentralized training is open biases… open biases means that everybody gets to contribute to reinforcing the model, and so therefore the biases are, in theory, supposed to be more balanced.” — 00:29:54

“Big companies like OpenAI and Microsoft and Google of the world are going and lobbying heavily and proposing that we ban open source AI because it’s in their best interest… they’re going out with the message saying that if you open source AI, China is going to steal our AI… scare tactics, classic — to make sure that they are the only ones… they’re playing this dirty game.” — 00:16:38

“The model is open source but the weights are not open source, and weights are available to the folks that contribute compute. So if I participate in a training run for [a] 100 billion parameter model, if I give my compute, I get some token that represents my ownership of the model… during inference time my ownership will give me some share of the profit… fully open source, owned by the people that built it.” — 00:26:46

“As you spend time on the Capitol Hill, you realize the forces that want to keep AI closed source are very, very active, working very hard to keep AI closed source, and they have incentive to do so.” — 00:59:39

“We humans physically stopped evolving about 10,000 years ago — our brains did not get bigger — but we got extremely smart… because we can coordinate and we can share knowledge… the most important thing for the evolution is open access to information. The more open and more frictionless there is a way to access information, the faster we can [evolve]… that’s why I think open source is not just beneficial to one single entity like Akash but beneficial to the whole ecosystem.” — 00:10:15

“Any sufficiently successful software has to be open source — be it Linux, be it Android, anything that you would deem will have impact on a globe. Internet is open source. But we saw a direction with the cloud that was opposite: it was getting increasingly closed, increasingly oligopolized by few players, and that was a future we didn’t want to accept.” — 00:00:48

“Amazon is offering these services that [are] essentially open source software white-labeled as Amazon services — that hurts the author of the open source software whose only business model is hosting… The way we’re going to do that is to allow the open source author… to offer their services directly to customers, cutting out the middleman… Open source services need an open source network… databases and caches and data warehousing services — you’ll probably see them in a year’s time on Akash.” — 00:25:45

“You’ll have a deep confidence in, like, open source AI that it doesn’t matter what centralized AI is going to produce, we’re going to beat them pretty, pretty aggressively, right? So, it’s like what happened with Linux… Look at where it is right now, 97% of the internet runs Linux.” — 00:43:45

“For the first time it’s so clear and so obvious and so evident that open technologies will exceed closed source technologies, and being closed source is not a moat at all, and communities will overpower companies.” — 00:08:31

“If you look at similar patterns, OpenAI is the modern AOL of the world. In fact I tweeted about it two years ago on this very exact fact: that any sufficiently successful technology has to be open source.” — 00:07:05

“It’s something that we absolutely need… it has to happen, because historically speaking, any sufficiently important technology that reached mass adoption is open source — Linux, worldwide web, you name it… Transformers, open source, right?” — 00:44:46

“Similarly, like OpenAI, I think, will be the new AOL of the world. Yes, they were the first to get to market with ChatGPT… that kicked off an entire industry, but I think the true intelligence through AI will be open source.” — 00:46:11

“We have to create an economy for open source contributors to be able to sustain themselves, to build open source software, and I think that is what Akash is going to transition to with the services economy, and that’s coming next year.” — 00:47:56

“We’re seeing open-source clones, or open-source GPT models, that are actually getting to be really, really good — as good as GPT-4 in some cases. And the time it’s taking from an OpenAI release to a clone is shortening really quickly.” — 00:21:32

“We were very, very scared because open source is keeping the lights on for hundreds of millions, thousands of developers right now. Because closed source is getting stronger and stronger. The only way to compete with that is through open source. And open source will give foundation to decentralization in terms of training and whatnot.” — 00:42:25

“Intelligence should be a commodity that is free like water and air we breathe. Being controlled by a company or group of individuals that serve their own best interest is not aligned with the interests of the world.” — 02:18:50

“Shouldn’t this be cheaper or free? Like, what happens when the Chinese labs stop producing state-of-the-art open source models?… Then how do you get your models? Well, you can get from Claude or OpenAI, but then you end up paying a lot of money. I think it will come a time when decentralization becomes very, very important.” — 00:34:45