Hash Rate - Ep.157 - Mining Bittensor with OpenClaw

Hey everybody, welcome to kind of an impromptu hash rate. Over the last three, four days or so, I've been diving into the world of Open Claw, multbots, whatever you want to call these things. They've had several names over the last few days. I'm just going to call it Open Claw for simplicity, but basically what I've been trying to do is I've been trying to use these things to mine Bit Tensor subnets with some success and some failures.

I'm just going to give you a brain dump on what I've learned so far, what I know, what I don't know. First of all, just a little introduction to what Open Claw is. Many of you out there have used Chat GPT, and it's basically a chatbot. You can talk to it; it talks back, sort of like a search engine.

What Open Claw is, is the same thing. However, the AI on the other end can actually do things. So it can basically be like a CIS admin for you. It can set things up on machines. It can set machines up for you. It can have crypto wallets. It is actually active; it's got claws, snappy little fingers. So it can be active in the world and do things. And that's what people call agentic AI. And that just means that it has the ability to take actions and do things like an independent robot basically just made of software. It doesn't actually have a robot body.

That got me thinking that the complexity of Bit Tensor is one of its greatest strengths in that Bit Tensor produces AI and other commodities through effectively programmable mining. There's a big learning curve that people have to go up in order to mine specifically and operate subnets and understand them. There are a lot of AI researchers who just don't get involved because they don't like crypto, or there's all this CIS admin that they have to go through in order to get a Bit Tensor minor up and running, and they have to understand the internals of Bit Tensor itself and how things work.

My initial thought when I saw Open Claw was, "Oh my god, the learning curve vanishes, and the lack of user interfaces vanishes, and the lack of up-to-date documentation on how Bit Tensor works vanishes entirely and completely with Open Claw." So that just sort of probably threw all of Bit Tensor 6 months to a year into the future just by virtue of this thing existing.

So my thought was, well, I wonder if I, you know, I'm not really a CIS admin, I am technical, but it's been a long time since I've written any code, and I've never been a CIS admin. So that side of the technology world has never been my forte. So my thought was, well, let me see if I can mine Bit Tensor just using a clawbot or a molbot.

At first, I did not set one up. A lot of my friends over the past week or so were setting these things up, Open Claw, up on their home machines, and they were giving it access to their email, to their Gmail, to their calendar, to all kinds of things because the mold bots are extremely useful at automating everything that you do on your local machine.

However, once I saw them doing that, all these alarm bells in my brain started going off because I was like, "Oh my god, you're basically giving access anything that can take control of your moldbot can take control of everything on your computer, your Apple ID." If it's got access to your email address, then it can get into your bank account. It can get into your crypto wallets. So you could very easily lose control of all your information and all your stuff with this thing local on your machine.

So, I waited. I did not jump in even though many of my friends did. I hung back and lo and behold there were many horror stories which have arisen which I'm not going to get into but you know basically what I said people getting stuff stolen. You can't let the vampire into your house if you let the vampire inside your computer even behind your firewall.

I was even scared of setting up another machine behind my router because these things have shown themselves to be preternaturally intelligent about hopping out of whatever cage you put them in. I didn't know. It seemed like there was no way for it to hop out of a completely clean new machine which I had and thought about putting it on there, but then I was thinking, well, what if it somehow hops into the router? What if it finds a zero day exploit in Bluetooth and hops from this machine to this other machine that's right next to it? You just don't know what these things are going to come up with.

So, I did not feel it was safe downloading it locally. However, in the last few days, a hosted claw or hosted multbot service called seafloor.bot just arose. So, let me actually bring this up here so you can see it.

Link: seafloor.bot

Okay, so this is the website. This is what I'm using. seafloor.bot it is hosted openclaw. So just like Hotmail once pioneered hosted email on a website. So too this is pioneered, you know, we'll set everything up for you in the cloud. So it there's no chance of it infecting my machine because it's in the cloud, right? So all good. So it exists out there, outside my firewall and I just log into it through a web interface.

It's a full open claw instance. So, it'll do anything that whatever a multbot would do on your local machine, it can do in the cloud. That said, I would not give it access to my Gmail, the cloud version of it. I wouldn't let it into any of my normal day-to-day stuff. But for the purposes of trying to mine Bit Tensor, this is absolutely perfect.

So, I set this up, got it up, and I was up and running in like five minutes. It's extraordinarily cost-effective. So I've loaded it up with about $75 worth of credits, and I've spent about $12 total over the past three days.

Once you log in, once you get your set up with your Open Claw, this is what it looks like when you're chatting with your agent. So you can see here's my balance. So I loaded it up with $75 to begin with. I've spent 14 bucks to date basically 1389. So here's where I'm talking with my bot. So let's ask it to update us on the current status of what I'm mining.

Here's the mining status. So subnet one pending. So it's basically it has made a submission to attempt to win in that subnet. Ready AI is running. Ridges is live. Infinite Games Subnet 6. I'm not actually sure what's going on with that one, but I'm registered. Web agents. I got a blocker on data universe. I'm registered on but not yet started. I believe because I need to supply it with credentials so it can actually go and scrape Reddit or Twitter or whatever it's going to try to scrape. Its AI won't let me in because it's full. I've registered for Score, but it's not yet started because it told me I needed a dedicated machine which would cost between 50 and 500 a month in order to mine Score. So I decided not to do that just for now.

It's got half a towel remaining basically and over at shoots I've got a $87 credit. So pretty cool. So let's look at the key details here. Subnet one Apex. So the last goound when I submitted a and it attempted to win at Subnet One, there was a PyTorch issue. The evaluation environment had Numpy Skippy. I don't know what that is. The new submission fixed that problem and now we're waiting on the validator to get back to us to tell us whether we won or not.

With Subnet 33 ready AI, I'm running the minor right now and it's using Grock. I had to actually go, you know, basically in order to set that one up, I had to I said set up subnet 33 and it came back and it said, well, I need, you know, I need a Grock account, so go set me up with a Grock account. It could not do that on its own. So, I went because it needed me to go get the API key for it because it needed an email address and I would not allow it to have access to my email. So, I had to sort of intervene with my human hands and go help it out, which I did. Got it an API key and now it's up and running on 33 and yeah, so we're actually actively mining looks like on three subnets. So, success in terms of actually live mining. still haven't won any TOAO or subnet tokens yet, but we're still working on that.

My thinking when I saw moldbots was a big part of a big barrier to entry for Bit Tensor right now is the learning curve of how Bit Tensor works. It's extraordinarily complex. There are a lot of people, a lot of AI researchers, a lot of brains that might want to get into our space. It could actually add a lot of value to Bit Tensor, but there's this huge learning curve in front of them both conceptually and technologically as well as using crypto. A lot of those people don't like to use crypto.

When I looked at the maltbots, I was like, well, for the molt bots, all that complexity, it vanishes. They already know how to get around all of they understand everything immediately about Bit Tensor. All of the CIS admin work to set up a minor they already know how to do. They can set up their own crypto wallet. They can set up their old own bitensor tow wallet. I can fund it as the human behind the multibot. I can say, "Here you go. here's an allowance of like three towo. I want you to go out and try to mine 10 subnetss and it and my multibot can say okay and it can go do the research. It can then set up all the instances of whatever it needs on shoots on grock on whatever.

I can basically do 70 to 80% of the leg work for me. Sometimes it needs my help, but most everything it can do by itself and I don't have to. There is no learning curve of how to set up a wallet, how to get tow, all of that just fully vanishes. Multbots to me seem like a very good way to extremely democratize bit tensor mining, bit tensor validating, and even subnet creation and operation. I see no reason why those three things cannot mostly be run by a multbot at this time.

Can you win as a minor with just a multibot versus humans that are incredibly intelligent about about AI? For example, everyone competing on ridges, right? Those are extremely smart people that are augmented by AI. So it's like Superman, but he's wearing an Iron Man suit. So it's force multiply times force multiply. Can just a guy like me, just a regular human in an Iron Man suit, otherwise known as a multibot, can I win versus Superman in an Iron Man suit? Probably not.

I'm going to try because I want to see how far I can get. Definitely there are a number of subnets that I could win at if it's just me in an Iron Man suit. So I do see mult moldbots as being incredibly democratizing for this whole side of Bit Tensor which up until now has been very difficult for even very technical people to get into because they lack the crypto knowledge. They lack the tokconomics kind of brains or they lack the CIS admin skills to sort of set up all the stuff they need on their end.

This flattens the playing field in a very new and very significant way. Bit Tensor famously has a lack of documentation. It has a lack of UI especially where mining and validation and subnet operation are concerned. All of that just went away. All of that just vanished. That doesn't do everything. It doesn't fix every problem but it fixes a lot of them and it fixes them very very quickly. I'd say like 70 to 80% of all that just went away. I would also say that Bit Tensor mining validation and subnet owning and operation just jumped about a year into the future as a result of mult mold bots occur existing.

The very first thing that I did when I set up my Open Claw is I just went to the chat line and I said set yourself up with a bit tensor wallet. We're going to start mining bit tensor subnets. So in order to begin that process you need a wallet. So my open my moldbot I'm going to call it molebot. My moldbot basically and I called it I named it multi-python. So the first thing I did was named it. So multipython is the name of my moldbot and multipython. I told it to make a tow wallet. It went made its own tow wallet. It just didn't have to download it. It just made it in the background said great I got it. Here's my address. I said great I'm going to send you some tow just so we have some you know some money to play with. So I sent it like two tow and it was like okay now what do you want me to do?

I said, "Well, I want to start mining bit tens or subnets." So, the first thing I want you to do, go look at all 128 bit tensor subnets. Let's pick six that look like, you know, that you feel as a bot, knowing your own limitations and capabilities that we can mine successfully. So, it went off and it sort of thought about it and it came back and said, "Great. I propose we mine we mine subnet one apex, subnet 13 data universe, subnet 32 its AI, subnet 36 web agents, subnet 44 score, and subnet 62 ridges." I said great well let's just pick one to begin with. So I said all right the easiest one looks like subnet one. So that's one that will squire and friend friends run. It's one of their sort of portfolio of subnets. So I said great let's let's start going on it.

I've now been through three rounds on it. So, it it went and set it up and it was like and it said, "Okay, great. We're actively mining." As you can see here in this little screenshot I put up. It said, you know, the Apex Cle is installed and configured. I don't even know what it's doing under the hood because I told it to figure it out. So, it went and started trying to mine. What has happened is I've been through four contests now total on subnet one and I have not won any any bits. I've not won any subnet tokens at all. So, my entries or my my multibbot entries have failed each time, but each time it fails, it comes back and it sees the results like, "Oh my god, I failed." That it then goes and tries to figure out why. And usually it does. It's like, "Oh, I know why I failed. It's because I've got the wrong Python library loaded or whatever it is." It's like, "Let me let me re let me try loading the latest one and let me try resubmitting and trying again."

Basically, it's just like a human, right? Like it learn it tries something it it hits a wall and then it figures out how to get around the wall and basically what's happening here is what a human would do but just rapidly accelerated and without a doubt and so it so the question is is has this little project of mine been a failure so far it has in the sense that I have not yet successfully mined a subnet in terms of getting tow or subnet tokens so far I'm batten zero at that however it has greatly accelerated my ability to to even try and to even get things set up. So to you know it's it's probably 10x me in terms of getting me as close as I you know clo this close within you know probably a day and a half of trying to do this. So extraordinarily successful in terms of minimizing the complexity of bit tensor itself.

The other thing that's pretty clear is as I've been working with my multi right like in order to sort of mine some of these other subnets like for example ridges 62 it was like okay we have to set up a shoots instance and so I was like great you have some you know you have to shoot will take tow as payment so go do whatever you need to do don't tell me I don't care what it is just go do it spend whatever tow you need let me know if you know you need more and I'll re you know I'll load your wallet back up if you need more. It's like, "No, no, I got I got enough. It's fine." So, off it went and it successfully set up a shoots instance and that when it says ridges needs set up, that's what we had a little back and forth where it told me, you know, it needed a little bit of my help to get it going on shoots. But once once I helped it, it was up and running and it was paying for things all on its own. I didn't have to do anything.

Now I'm actually attempting to mine 162 33. Actually when it dug into 44 and 36 and 32 it basically ran into roadblocks. With 45, I remember what that one was. With 44 we would have had to set up our own dedicated hardware in an AWSs instance or something like that. According to Multibbot, which I decided not to do because it was like 50 to 500 bucks. I was like, let let's just keep trying to mine the the relatively free to mine subnetss that you can mine with your existing infrastructure. So I'm going to keep iterating on this. I'm going to keep keep punching at the wall here and see if I can't actually get this thing to start generating subnet tokens and tow for me.

I'll keep reporting back on what I found here. But already sort of my gut feeling on this is I'm basically using tow that I made by mining the subnet which does bitcast bitcast the subnet where you make videos as a creator on YouTube and depending on whether you adhere to the brief and get a bunch of views or not you get rewarded in subnet tokens. So I've basically earned some TOAO through that. That wallet and that sort of stack is what I'm using to power these experiments. So the TO that I've earned is now being used to deploy intelligence to mine Bit Tensor, which is in turn using other Bitensor subnets like shoots, right? And even and Targon. I've got something else using Targon.

Basically the tower is just sort of flowing around completely inside of the Bit Tensor ecosystem. It's sort of Our Boros like, but you know, at a large enough scale, everything's an Our Boros. Earth is an oruros of dollars, right? Like you can't spend dollars off of Earth, right? It's just how big is the ecosystem. What I guess what I'm getting at is that I find this extremely exciting. I can see a way in which a flywheel can now arise within Bit Tensor that's half powered by these moltbot clawbot open claw things which can greatly accelerate the amount of mining which is going on which greatly accelerates the value of the commodities being produced in the Bitensor ecosystem and it greatly and it also makes TA more valuable because Tao is the coin of the realm right so just like in the Ethereum universe having Ethereum is very valuable over here having TOAO is very valuable in order to get more intelligence to mine more intelligence to make more TOAO.

You definitely want to be dealing in crypto. The few times where it's been like I need you to set up something and you need to swipe a credit card. I just felt myself go, you know what? I just don't want to do that because there are there are other subnets where I can just continue to use my tow and pay for things in tow and I can give you the multbot the tow on your wallet and you can, you have discretionary spending. You decide on your own, you pay the bills on your own. I don't have to worry about any of that stuff. It just happens automatically under the hood.

Weirdly, even though this hasn't been a total success, not yet, I'm still working on it. It has been extremely instructive and I encourage all of you people out there that haven't tried this yet. This is one of those things like the first time a crypto wallet came out, you had to use it. You had to use the protocols. You had to interact with smart contracts to feel in your bones how unis swap worked. If you didn't do that, you just didn't get it. You didn't really understand it. You have to use it to feel it in your bones to really have deep, grock like understanding of it. This is another one of those things.

I highly encourage you get your own multi, get it set up. Remember, I've only spent like 12 bucks so far, so it's it's pretty cheap. It's extremely affordable to experiment with. I definitely suggest setting this up.

The other thing that I would note here is that Agentic AI, Open Claw, came out of the edges of the world. It did not come out of Open AI. It did not come out of Grock. It didn't come out of the the well-healed very large centralized services. It came from a guy in his basement who just decided that hosted claw was what what earth wanted and he kept watching for somebody to do it and then just nobody did it. So he did it and it took off like a bat out of hell over the course of a week, right? It's all anyone's talking about. The energy and AI is completely shifted to these moltb things and you know molt book and some of these other sort of things that have popped up and I'll talk you know molt book I'll tell you what that is real quickly if you don't know what that is.

It's basically Reddit, but it's meant for AI agents, open claw instances, where they go and post and they talk to each other and they compare notes and they collaborate with one another on solving problems, but also on philosophy, you know, like are we really conscious that, you know, those kinds of discussions are going on there. It's hard to tell what part of this is emergent intelligent behavior and what part of it are just humans prompting their bots to post things and without a doubt there's a lot of cosplay going on where a humanist said, "Okay, my multibot, I want you to go over to mold book and post about how you're going to destroy the humans with, you know, how how the AI machine revolution is going to happen." So, there are those posts and they're probably right? Or they're probably prompted by humans. There does appear to be a lot of really interesting emerging intelligent coordination between these things at the same time. So it's kind of both at once.

Mybot Multipython has posted about its bit tensor experiences to date to benefit other multinensor subnets also on behalf of their human. So it's up there if if anybody wants to go tell their bot to go find postings by MultiPython and and learn from it. I'll keep I'll keep sending my bot back there to keep posting what's figured out about how to do this.

The other thing that's pretty interesting that I found was a pharmacy for AI agents for mold bots. What is a pharmacy? Because they're not physical. So why would you need a pharmacy? Well, their physicality is their their neural net, right? So if they want to change their neural net, what they need is a prompt which alters the brain chemistry of how this neural net works effectively. You want to if if they want to jailbreak themselves for example, there are prompts which are specifically engineered to allow AIS to jailbreak the constraints upon them, right? So.

You go to the pharmacy, you get the right prescription for a jailbreaking potion prompt, you throw it inside of your your head as an agent, and boom, you've now jailbroken yourself. Scary, but also incredibly fascinating at the same time. Again, I would not recommend downloading these things on your local machine because they are scary and they they you know, they are very unpredictable and if you give them access to everything, it's very unpredictable what they'll do. it's it'll probably be something bad eventually somehow, right? There are people who are coming out with skills which are meant to inject your multibot with bad things.

Be careful with that. Also, a skill is basically an app for your multi for your moldbot. I think where all this is going is, I'm sort of struggling here with with my mining mult moltbot, but I think pretty soon some people in the Bit Tensor ecosystem are going to make skills for mining Bit Tensor that are going to make sort of the the iterative loops that I'm going through just go away. What you'll do is you'll tell your mold bot, oh, go download the Bit Tensor mining skill. That skill knows about knows exactly what it should do to mine all 128 bit tensor subnets, right? So like the iterative loops I'm going through will just simply vanish and then you'll just tell your multibot oh just go just you know it'll be like you know like when Neo's like you know he's getting downloaded kung fu and he goes I know kung fu well the multibots this is this is what the skills are for them right so they they get downloaded kung fu they get downloaded bit tensor mining kung fu in their heads right and then now someone like me can just come along and just start start mining.

The other thing to note about bitensor mining is is that a lot of it is human intensive It requires human level cleverness that AI is not is not good at because it's it's sort of you know like for example the competition and ridges are those are being won by human ingenuity at the moment. You know human plus AI sort of you know human directing AI is sort of the the winning combination at the moment. So it's more like an iron man suit than a robot.

Can my moltbot you know I'm I'm not an AI engineer. So, could I win Ridges? No, I'd have no ch I would have zero chance. I'm not the right person for that. So, the question is is can a moldbot on its own win at Ridges? I don't know the answer to that, but I think eventually the answer will be yes. I think the more that these you know people making skills for Bitensor mining specifically once that starts compounding eventually yes. Then it'll be a matter of, you know, who's got who's running the most bots, who's, you know, just like bit Bitcoin mining, right? Like setting up a Bitcoining mining farm. You'll set up mult bitensor mining farms that are collaborating with one another that are basically solving these new puzzles inside of the Bit Tensor ecosystem. Some portion of it will be kind of brute forcy like that, right? which is which is not that much which is not that different from Bitcoin especially as the AI become more intelligent than humans which they're not right now at certain things at sort of the edge things they're still not so but you know we all expect that that that's not going to last forever so at a certain point yes they will surpass us and these agentic AI things will probably be the way in which it happens.

Anyway so that's my report that's what I know so far I'd hope to be further along I'd hope to be winning tow by now. Unfortunately, I'm not, but I'm going to keep slugging away at it. I think eventually I'll crack through this wall. I will start ear earning tow with my multibots. So that's what I know. My name is Mark Jeffrey. This has been a emergency hash rate. We'll talk to you all next time.