This feels illegal...

0:00 So I was scrolling X and I came across this project Dark Bloom. Basically you install Dark Bloom on your system and then all of a sudden you're sharing some of your compute with other people. You

0:11 can kind of think of it as a massive distributed data center and it is fascinating. There's a lot of people that don't like data centers and I've been seeing that opinion around and this

0:24 right here might actually be the solution. a completely distributed data center run by you. I think a lot of people got excited about it because you can make money with it. But that's not

0:36 what was exciting to me. I am super interested in this concept that we have a distributed data center. Rather than having these massive data centers which are highly unpopular, we can basically

0:49 break it up and they can live in everybody's house. Kind of like solar installations. Everybody's generating their own electricity. And here's how you check how much you can make. Here's

0:59 the earning estimate. It's dark bloom.dev. Click right here. Let's say we have a Mac Studio. We have an M5 Ultra with 96 gigs of memory. We can be making $37 a month. I know that doesn't

1:14 sound like a lot, but it's basically free money. There is a very small incremental cost for the electricity, but otherwise your computer, your Mac is just sitting there being not used. So,

1:29 it serves open-source models. Here are some of the models it's serving. Quen 3.6, Gemma 4, and GPT OSS. These are great smaller models that anybody can run and use. And when you aggregate all

1:43 of these nodes together, you get a lot of inference. And this project has been blowing up. It's only a few days old. So, I'm going to tell you all about it. Then, I'm going to show you how to

1:54 install it. This is so cool. Now, Dark Bloom is currently being served on Open Router. So, if you want to use one of the models provided by Dark Bloom, you can right now. It's really nice and it's

2:07 like 50% cheaper than other providers on Open Router. And they've already served 4.5 billion tokens in the matter of like a week. So, typically a data center looks like this. A massive building

2:22 placed in usually the middle of nowhere with hundreds of thousands of GPUs. And if you've been following the news lately, they are not that popular in the United States. This might actually be a

2:34 decent alternative to that. So, if you're one of those people who are super concerned about data centers, but you still want to explore artificial intelligence, this is a great option.

2:46 Obviously, local AI has been a thing for a while. You literally don't need to use any of these massive data centers or the power grid. You could just get it up and running on any machine. But now, imagine

2:58 this. Imagine you don't necessarily have a powerful enough computer sitting on your desk to run some of these great models. Well, now you don't need it. You can connect to somebody else. This is

3:11 peer-to-peer inference, peer-to-peer AI. Super exciting because now you're not using those massive data centers. You're just connecting to somebody else. You are

3:21 almost just paying them directly. And this is a very different paradigm than what we're used to. This actually might solve some of at least my concerns about the concentration of power in AI

3:33 open-source openw weights models being run on distributed GPUs really distributed compute. And for those of you who are wondering does this actually work and how does it work? They actually

3:45 put together an entire white paper about it. So the core problem, a user wants to run an AI model on someone else's Mac, but the machine's owner who has root access and physical custody should not

3:56 be able to see the user's prompts or the model's responses. So if you were concerned about this model and shipping your data to somebody else, which you know people are concerned about giving

4:07 the data to the big data centers or the open AIs and anthropics of the world, they have actually solved that problem with this project. Our approach eliminates every software path through

4:18 which inference data could be observed. The inference engine runs directly inside a single hardened Swift process. No subprocesses, no local server, no interprocess communication using MLX

4:31 Swift LM, which is Apple's inference engine on the Apple silicon GPU. So, I'm not going to get too far into the technical details, but they basically solve the problem of sending your data

4:42 to another computer and that other computer being able to observe it. Now, they can't. So, just imagine we have a million of these nodes set up. That is major inference power that anybody can

4:54 tap into at a fraction of the price of going to an open AI or an anthropic. And it's completely distributed. It is completely built by regular people. It's awesome. Okay, so let me show you how to

5:06 install it now. So the first thing you're going to do is go to dark bloom.dev. Click up here and click earn. And it currently has a CLI, although the Mac app is coming soon. And it's really

5:17 just this one command. You will need a Dark Bloom account. You will need to install something on your Mac, but after that, it's pretty much just sitting in the background making you money. And to

5:29 be clear, this is not an ad for Dark Bloom. I make no money from promoting Dark Bloom. It is just such a cool project that I wanted to tell you all about. So, you know what I actually did?

5:40 I just went to codeex and said install and set up Dark Bloom AI and it's going to do everything else for me. To finish, I have to just sign in. So, done. Enter the code shown in terminal. And again,

5:50 this is all being done through Codeex's own browser. Okay. So, then this thing pops up in Mac OS and it's under device management and settings and it's Dark Bloom provider enrollment. And all you

6:01 have to do is click enroll. And that's pretty much it. And look at that. It's even testing how many tokens per second I'm getting. And I'm getting 78 tokens per second. So now somebody else can use

6:11 this these tokens and pay me for it. It's really awesome. Okay, so it tested it's running GPT OSS 20B. We passed at 78 tokens per second, but now we have to do hardware verification. So I'm just

6:25 going to say finish it for me. And here's something you should know. Right now, they just made this change where you need at least 48 GB of RAM. And I know that's a lot, but most Mac minis,

6:37 most Mac Studios come with at least that much. So, you should be good. A lot of the higherend Mac laptops also come with that, and they'll probably reduce that size as it goes, but they raised it just

6:49 because they were getting so much demand, they had to have some kind of quality bar, and that's what they decided on. Now, the way that you get paid is by connecting a Stripe account

6:58 to your bank. Now, Stripe is the intermediary. So, you're not connecting it directly to Dark Bloom. They have no access to your bank account. They can only send you money. So, obviously, use

7:08 it at your own risk, but based on what I've seen about it, nothing screams scam to me. Now, I made a video and I put it on Instagram yesterday and a lot of the comments said this is a scam. like a lot

7:21 of them surprisingly but the software is publicly available. You can go inspect it. You can go audit it. It is publicly readable. So I actually had GPT 5.6 Soul review the entire codebase because I

7:36 wanted to know is there anything malicious in it. So the only thing it really pointed out as a problem is that currently you get 100% of the revenue earned but that can be changed in the

7:46 future. Okay, kind of figured that. There are no hidden malware, no mining, no credential theft located in there. MDM and device access concern is low. And so again, use it at your own risk,

8:01 but it seems benign to me. And they also put together an entire white paper about how the tech actually works. So it's quite transparent right now. Now again, it is in the earliest of early days, so

8:16 there's going to be rough edges to it. There's going to be things that might not work as expected. They're going to adjust things pretty rapidly. But why I wanted to share this is because it seems

8:27 like a really cool, interesting project and more importantly, a very compelling concept. There's so many powerful computers out there that are just sitting idle and we're all dependent on

8:38 these massive data centers. Own your own compute. Now the other criticism from comments seems to be about electricity usage and I can tell you Apple computers are extremely efficient. So the

8:53 incremental cost of that electricity is going to be very low. Now obviously it depends on where you live, what your electricity costs are, but you can probably use AI to do that calculation.

9:04 And if you're earning, you know, a hundred a couple hundred bucks a month, your electricity bill is not going to be more than that. And again, this is all very experimental right now. I just

9:14 think it's interesting. And if you want to stay uptodate on the latest in AI, check out our newsletter, forwardfuture.com, link down below. So, I haven't started

9:24 running it. I haven't gotten paid yet. I have installed it and I am waiting for it to kick on and start serving, but I'm looking forward to it. And it's not about the money to me. It's not a ton of

9:35 money, but I do love this idea of distributed compute. It is very exciting to me as an alternative to these massive concentrated data centers. And one last thing, this is also another reason why

9:50 open-source open weights models are so important. It puts the power into each individual's hands. You can do whatever you want with the model, including a cool project like this. and distributed

10:03 compute feels like a very real thing and something that can be very powerful in the future. And so I wanted to explore it. I wanted to test it. We're in the very earliest of innings, but that's why

10:16 we're here. And I know that's why you're watching this video. You want to be on the cutting edge, on the frontier with me. And I made an entire video about open source. Check that out right