Doug Sillard: Taostats, Bittensor Dynamic TAO, Chain Buys, MEV Bots & TaoFlow Explained | Ep. 82

When he explained to me how Bit Tensor worked and the fact that all of these subnets are actually companies creating products that are useful to people, that sounded interesting to me.

All of a sudden we've got, when I started there were 32 subnets, now there's 128 subnets and each one of them has their own niche of what they're trying to create and build, and to me that was just fascinating.

Like here we have this sort of like mini ecosystem of startups all getting powered by bit tensor. It's sort of that balance. And I think right now if you asked the people who look at all the subnets, so there's still enough subnets that just aren't producing the the the best products that they could.

So like as soon as we start getting a real good mix of subnets, then maybe they'll say, "Okay, well maybe we'll add another 20 or 30 and then we can start growing," because the subnets are getting deregistered now.

There's been very little outcry when they get, I mean, there's been some like I can't believe that one got that they have such a great idea, but there are companies out there that have great ideas that don't make it just because they weren't it wasn't built in a way that made sense or you know there was other problems with the company.

And that's what happens with any startup, 90% of startups fail so you know that sort of thing is probably going to happen in bit sensor.

I do also think it means that as a subnet you've got to hit the ground running. When I started there was subnets that took three, four months before they even announced what the heck they were doing.

And right now, as soon as you're as soon as you launch, as soon as you start allowing a mission, you've got to have a project and you've got to go. So, I think in some ways that's really good.

We don't have these subnets that are sort of like, we're not really sure what we're going to do yet. Like, you have to have a business plan and go from the get-go.

Welcome to the Ventura Labs podcast.

Today, we are joined by Doug Sillers, Talstad's Devril expert. He shares his journey from aspiring professor to entering crypto via bit tensor in 2024. Doug discusses challenges like mebots, tflow dynamics for new subnets, chain buys, and evolving admissions mechanisms and explains complex concepts like root claims ADR and the balance between decentralization and necessary interventions.

Now back to Doug with the 82nd Ventura Labs podcast. Doug Sillards of Talstats. Yeah. Mr. Deell, how are you doing today?

Doing good. I'm doing good.

What is your official role at Talstats? Is it Devril?

It's a good question. Um, every time I'm asked to fill it in on like a website, like what I do, I change it. Um, I don't know. Uh we're a small group so it doesn't really matter.

I do developer relations but I'm also doing a lot of community management. Most of the people I'm talking to aren't really developers. They're just people who are interested in in Tao and Bit Tensor.

So am I helping developers? Yes. But am I helping the community? Also yes. Am I writing a lot of educational docs and creating videos? Also yes. So it's sort of you know whatever needs to get done.

How often do you find yourself reanswering the same questions daily?

I try to reach back and I think this sometimes goes back to just having been a parent. It's just like you get asked 500 questions a day and when you're doing community management you get asked the same question five you know and like earlier today there was a blip on to stats and someone said hey is to stats acting up.

I'm like yeah we're working on it. And then the next person said, "Hey, is to stats down?" And it's like, "Just read one message. Just look up."

How do you find the patience sometimes to deal with those Discord messages where it's like, "Hey, if you would have just searched, you would have found the answer."

That's a good question. I try not to get there. There are moments where I'm just like, "Okay, deep breath. Don't be a smartass." Um, but there are also moments where it's Friday afternoon and I get a little silly and punchy with the answers.

But you know, generally it's people who want to know how bit tensor works and as you know it's complicated, right?

You know, I sort of think about it as like a lot of people know that in baseball there's three outs and there's nine innings, but most people don't know about the ground rule double or the infield fly rule or, you know, all these weird things that can happen during a baseball game.

And so, it's my job to be like the umpire. I need to know three strikes, you're out. But I also need to know that the infield fly rule only applies when there's certain runners on base with a certain number of outs, right? I need to know all of those rules.

Let's talk about what led you to this moment and working at Talstats and being at Bit Tensor. Your history before Bit Tensor was developer relations and you would go around speaking at all these different events. How did you get into that life? How was that your career at the time?

It was completely by accident. Um, I've been doing develop. I thought I wanted to be a professor and so I really like the idea of teaching. Um, I then went to grad school and realized I didn't want to be a professor.

Um, and I landed a job at AT&T where I was helping developers fix stuff and I realized that that's what I enjoyed doing and I've sort of continued with that role. um you know creating that educational material so that people can be successful at what they're trying to do.

Coming from that traditional traditional developer relations role and ecosystem, how did that translate over when you found Bit Tensor?

So some of my first videos that I created are still up there on the Tostats YouTube channel. Um I'd like to think I've gotten a lot better at presenting those videos. they're a lot more colorful and and I think um entertaining in a you know in an educational way.

So I think I'm presenting that better. I I just think coming in and taking the time to understand like what the heck is going on was really the most important part. Like it took months before I really knew how it all worked and felt comfortable explaining it.

So some of my first videos are a little awkward cuz I was still, you know, the best way to learn something is, I think, to teach it. And so I spent a lot of time just like figuring it out by creating videos that were mostly teaching me and then hopefully it helped other people as well.

And what was your introduction to Bit Tensor? Was it you found Bit Tensor or did Bit Tensor come find you?

Mog found me. Um, he was looking for somebody. This was back when Corell was still around um which was using subnets. It was subnets 18 and 19. It was all inference and it was a web page with an API.

And so he wanted docs on that but he also really wanted great docs and information about to stats and how bit tensor worked because he's I mean we have docs. there's the the the official docs um and TAS has docs and I think there's still room still today to do better because there's just so much going on in the ecosystem. It's hard to find all that information.

So, uh Mog found me because he thought found found that as something that was lacking in in Bit Tensor.

And then what was your first impression when Mog reached out to you to come docs for Corsel?

You know, a lot of the companies that first reached out to me who were, this is my first crypto job. Um, and a lot of the companies that had reached out to me before were like, we're building NFTTS or something. And I I just I couldn't really find like what's that real world application for whatever we're doing.

Um, but when he explained to me how Bit Tensor worked and the fact that all of these subnets are actually companies creating products that are useful to people like I that sounded interesting to me, right?

All of a sudden we've got, you know, when I started there were 32 subnets. Now there's 128 subnets and each one of them has their own niche of what they're trying to create and build. Um, and to me that was just fascinating, right? Like here we have this sort of like mini ecosystem of startups all getting powered by bit tensor.

So you came into bit tensor when there was 32 subnets. This was pre-dynamic tal. What was your understanding of hey this is what bit sensor is at the time when you first encountered it?

I think I kind of described it right you know um I mean I just had no idea how any of it worked. Like what's a minor? what's a validator, what's a subnet? Um, and I just spent a lot of time like trying to understand the vocabulary like what are we saying here? What does this mean?

Uh, Mog was patient with me when I asked questions. But, uh, just trying to understand like how it all worked. Um, you know, at the same time I was trying to understand corell which was just like it that API felt a lot like you know using the uh open AI API, right? Hey, I want this to happen and you know you know what's the fastest fish and it would reply back to sailfish or whatever, right?

So I was learning, you know, it was sort of just there was this fire hose of data coming at me and I sort of just had to, you know, compartmentalize it in a way that made sense to me. And so the first version of the docs uh made a lot of sense to me cuz I was compartmentalizing in a certain way and it sort of evolved over time uh to make more sense for people who are in crypto and um who understand bit tensor.

These docs that you write are not just for tal, you guys also support docs for bitensor generally is this correct?

Yeah so it's docs.talstats.io IO. And so there's a whole section, you know, there's an API. We've got I just this weekend I discovered we have 167 endpoints.

Um not 10, not 10. Not 10, not none. Um there are 167. Um and people say, well, what's in the API? And I say, you know, if you can find it on the website, and you know, there's a lot of data there. I get surprised on the website sometimes. I'm like, I didn't even know we had this. um it's in the API so you can build your own version of any of those tables if you have um certain need and I get people all the time why don't you add this to the table I'm like they're already 37 columns I we can't add anymore it's it's way too big um so here's the API you can build your own table and you can generate it whenever you want.

So you know that was sort of uh we have docs that describe how to read everything on to stats, but then we also why do we show these values and so we talk about emission and we talk about um you know what miners do and what validators do and what the numbers mean for miners and validators etc.

This might be really in the weeds a bit, but there's archive nodes and there's light nodes for sub sensor and your API serves an archive node and archive nodes I believe are like three terabytes worth of data. What is the complexity level of somebody trying to do this themselves versus just using your API and the cost perspective of this?

I have so running an archive node clearly you need a big machine right you need to have three terabytes of storage just to hold the archive node and they're getting bigger um light nodes are great for data in the last couple of hours so if you're only looking at like I need to know what's on the chain right now you can run a light node and you'll be fine.

What Talstat has done is we've taken all of the data out of those nodes and we've indexed them so we just have giant databases that hold all of that data in a way that you can pull it quickly because the nodes are a lot of the data are in those nodes, but it's not in a way that is easy to to grab.

So things like transactions, they're in there, but how do I know how do I find all the transactions for your wallet? You can't really do that with an archive node easily. But since we've looked at every single block and in indexed every single transaction, we know everything that your wallet has done in the history of Bit Tensor, it's in our database. So we can just select that wallet and pull back all those results for you.

Is your role at all building out the UI to display this data or is it we've got a team of front-end developers who do all of that. So I will come in there and I word smmith sometimes, right? The little mouseovers like sometimes I say, "Hey, we should change this cuz that's not quite right." Um, but for the most part, the front end devs are building it. We've got designers who are designing the pages. Um, and then we've got guys on the back end who are doing all the indexing and making sure it all runs.

How big is this Talstat's team now? This sounds I don't know how many there are to be honest, but there are a bunch of us. And uh what's really cool right now is subnet 19 is going to become the backend for ta stats. So we're we're launching block machine which is going to be doing all of the it's going to be all the nodes and then we can we can index based on that block machine.

How is the design for block machine different from what sub vortex used to be or something like flame wire that's currently live? Because there's been a few attempts now at these.

That's a great question and I am not the right guy to answer that question.

Okay. Um I I like listening to what they're talking about but I can't explain it to you.

Okay, that's fair. That's fair. I can't expect people to know other stuff that's like this. You are Mr. Mr. uh chain dev relations here. You had well you have a lot of videos on the TA stats YouTube page that explain a lot of the concepts for bit tensor pretty succinctly. One of those I watched earlier today was the chain buys.

So the chain by feature, my understanding of your explanation, you can correct my relay of information here, that is you take the EMA of the tflows and then you have the price of a subnet and you cap the EMA for emissions at that price and then the excess the difference between the tflow EMA and the price is what goes to chain buys if it's higher right?

Yep. Okay.

You can't what? Yeah. Go ahead.

Well, I was going to say, what do you think was the benefit of this feature, if any? Do you like this feature or is it a bad feature?

So, we've had this for a while because it if we can't emit more than one before the having you couldn't emit one more one more than one alpha into the pool. Now, you can't emit more than half of an alpha into the pool.

Um, and so what this does is if the emission is higher, it goes into a chain buy. And so what that does is it ends up because there's tow um going into the pool and then the alpha is being recycled, it actually causes the price to come up and so that the emission and the price sort of equalize.

So I think it's a good thing. I they had to do something with that excess with that extra um what we call excess tile like something had to be done with it and this seems to be a good way of dealing with it. Um there have been other talk there's been talk about doing other things with it but you know for now I think this is a great solution.

If we take a step back just dynamic tal in general, you're probably one of the people with the best understanding of the design and all the concepts and moving parts behind it. Do you think this was an optimal design for dynamic tile with alpha in alpha out each subnet having its own token or was there a way to simplify the premise of this?

I think what they did was a nice um compromise between um some of the ideas that they had were crazy complicated um and would have just the chain would have gotten really big. There have been a lot of chain bloat and it would have been very complicated to make it work properly. So by doing something sort of in between it was a nice um a nice compromise.

Was it required that every single um subnet gets their own alpha? Um I think it's played out pretty well. Um so I to me I think looking back at it a year later um because I guess it was a year ago like Friday um that Detail launched. So we're right here at a year. Um, and I think so far it's been a success.

Like there's some been some bumps in the roads, right? We've made some changes like the um XS tow. Um, you know, I don't think we're done with the way emission works on Bit Tensor, right? These things have been changing. I've been here two years. They've been changing constantly. We're not done yet, but I think the detail was a good way, you know, it was a good move push forward.

We're not done yet with the way admissions are changing. What are what are some of the rumblings you're hearing about what the future might look like?

Good question. I don't know if there's anything I mean there's there's some tweaks going on right now to like um stop emissions on subnets that are misbehaving. Um, so you know, again, these are sort of like those those rules that nobody really cares about unless they apply to you, right? They're really really inside baseball sort of uh complicated things to deal with.

Are there going to be big changes like DTO? I don't know, but I'm sure there's going to be some tweaks into how it all um you know, I think just in December, just you know, a couple months ago, they changed the way um root uh root staking worked, right? So now you can you can uh swap the way it always did. So if you stake into root, you get tow, but now you can keep so you get a little bit of alpha in 128 subnets. You know, that was just 3 months ago.

Yeah, it's true. We're only at a year of detail. And with the you kind of mentioned it there with the dregistrations, do you think the triumvirate or the Senate should be able to vote to dissolve a subnet?

There is talk about that. I it doesn't feel very decentralized, but I do see that there is there are times where that might be something that really needs to be done, right? like if a subnet is doing something that's just so terribly negative towards the chain or towards Bit Tensor.

Um but I don't think that's something that should be taken very lightly, right? Because we're supposed to be decentralized. It should be that the the people speak. Um and I do think that if a subnet is doing things there was, you know, the 67 did a rug pole and their price went way down and they were dregistered within a week.

So, I do think that there are mechanisms in place to take care of that, but if something needs to happen faster, I I don't see why it I don't see why it shouldn't be there, but it's definitely something that should only be done right like five alarm fire sort of uh fix.

What do you think of the way const had handled the Loca Submit 28 way back like in the beginning of Detail?

I'm trying to remember how it all worked and what they did, but I think they basically said the subnet, you know, they I'm trying to even do you remember what they did?

Yeah. So, he asked other validators to basically like child key stake to him and then he said the validator code's going to be we're going to reward the people that unstake the most alpha and then

Right. Yeah. So, he just changed the rules of the subnet.

Yeah. Exactly. again like something that should only be done, you know, when it's when something, you know, there's there's a couple subnets right now that um have been called out for not behaving well. And so that's where some of these fixes are coming in to sort of quell that behavior from ever happening again.

Um, you know, I think that's good. You know, I'm sure we're going to talk about Mevbots at some point along here, right? There's there's always going to be in any society people who find the the ways to sneak through the cracks and to you know capitalize for their own financial advantage. So um anything we can do to sort of like you know plug those holes as as the chain evolves is a good thing.

That is a good point. Bit tensor is built on incentives here and mebbots are are just chasing the incentive. Why is that such a hard problem to solve?

I don't think it's so looking at the data and it's been a couple weeks since I looked at it. Um the first So what's happening with the Mevbot first off is you do a stake and if somebody enters a uh stake right in front of you, the price changes.

And so what's happening with these Mevbots is they put a bunch of towing in front of you. You stake and the price changes a whole bunch and then they unstake and you get a high slippage and they make the profits off of you.

And so the first thing that the chain did is they applied limits. So you could say I'm only going to let my price go down by 1%. Or half a percent. And that really did stop a lot of the meving.

The problem is um a lot of people are writing their own scripts and they're not using that new extrinsic. So they're still using the one that leaves them wide open, right? So it's sort of, you know, if you don't if you set a strong limit and you do a transaction, you're not going to get me or you'll get me for a tiny fraction of, you know, how much you did.

But there are people out there who are not using those limit extrinsics. the mev bots are attacking them. They're people who are doing the with the limit and they're like 10%. Well, that's too big. You're going to get me.

Um, so the idea of Mev Shield is to encrypt that so it's harder for the Mevbots to see it. And of course, they're figuring out ways that they can sort of see it. Um, so I mean the real solution is if you're staking and you have a really tight limit, um, you're not going to get me and you'll be fine.

But the problem is a lot of people aren't doing that. So the chain and the devs are trying to figure out all these different ways to protect people from from themselves to I mean they're protecting themselves from bbots, but they're protecting them from themselves by not using the right tooling to do it correctly.

Another kind of thing in the same topic space here was the V3 liquidity pools that was in bit tensor and then went away after talflow happened, right? Why did it go away? Do you think that comes back? Is it needed? What's the benefit of it?

There are I've seen uh some ideas floated on how to bring it back. Um I think people were using the liquidity pools in a way that was harming the new subnets. So I you know new subnets struggle with tow flow. You need to have your flow stay remain positive to gain emission and so the more towel that comes in um the higher your emission.

But you have miners who are selling right. So there's this constant, you know, buy sell pressure and I believe um I heard rumblings. I never really investigated and proved it to myself that people were using those liquidity pools to actually harm the tow flow of those new subnets and it was hurting that whole there's a whole liquidity issue with the new subnet and tow flow and I think those pools were causing an issue.

So they they said uh it's this design isn't perfect so we're going to come back with something better and then TF flow in the new subnets. Do you feel the design currently is unfair to new subnets?

So I you know I've looked at this and if you are a new subnet you need to in to get to get tow emission into your liquidity pool which is a lot harder especially after the having which we had in December. Um, you need to have tile coming in to get a mission. You need a positive tile flow to have that happen.

And I think you just need to part of your strategy when you have a new subnet is to know that you have to have just a kind of a a little bit of a backlog of tow that flow going in. You need to keep invest DC dcaing in tow to keep a positive towel flow. um over time um it's almost like you know it costs a 100 tow to buy a subnet but you need to have another 100 or 200 tow slowly keep that towel flow going until you hit that momentum.

I do also think it means that as a subnet you've got to hit the ground running. You know when I started there was subnets that took three four months before they even announced what the heck they were doing. And right now if as soon as you're as soon as you launch, as soon as you start, you know, allowing a mission, you've got to have a project and you've got to go.

So, I think in some ways that's really good. We don't have these subnets that are sort of like we're not really sure what we're going to do yet. Like you have to have a business plan and go from the get-go.

For Talflow, what are all the things that count and don't count? Like minor registrations, do those count for TFlow? the root prop does that count against like

No, neither of those do. I believe it's only staking and unstaking. Um, and that's where the liquidity pools, they weren't stakes and they weren't unstakes, but they were affecting the amount of TOA and alpha in the liquidity pool. So, that's sort of why they had to do something there with the V3 liquidity pools.

So, staking doesn't root prop does not. Um there was when um the keep and swap came out uh end of last year, there was talk of making root prop actually count as part of that. Um I think that's been shelved. I haven't heard anything about that, but there was going to be a another way that you could do your root um staking and the validators were going to have a part of it and it was going to affect um the tow flow. But I think that's been sheld for now.

But again, you know, like these emission changes happen and so it may come back later this year. What do you think about ADR? When ADR falls below one, do you view that as a problem?

So ADR and forgive me, but we've had a couple different definitions of ADR when it first came out. Greater than one meant that you got more tow than the alpha was actually worth. Is that the way you're thinking of it or are you thinking it

ADR I'm when I say it I refer to alpha outside the pool versus alpha inside the pool. That's what I view to and then okay the one is where they're equal. And so if if a subnet were to be dregistered if the ADR is less than one does that mean that the holder gets more tile than it was actually worth?

Okay. So that's I think the new definition, but I get confused because they changed the definition in the middle of the whole and it's like we're going to take the reciprocal and then everybody I don't know. Um so you know if it's less than one I think that's you know that's good, right? Your your investment is protected. Um I don't really have a strong feeling about it though.

That's fair. That's fair. There's too many things to care about everything. Yeah. And you're someone that gets a lot of questions. probably the most if not mach is the only person ahead of you for questions. What are the questions you see most often? What is the most common misunderstanding you're running into?

Right now it's people who don't understand that how root claim works because up until December you would see your Taiwan route you would see it increase every 72 minutes and they're like it hasn't increased. It hasn't increased for 3 days. Why is that? And it's because they change the with the new keep and swap. It now happens on average every 2 days.

Um I went into the data. It's about every 36 hours on average. So it's actually less than 2 days. But you know, as you know, with any sort of distribution, if the average is 36, there were outliers where it took 10 days. And so I mean it's statistics, right? Like if you have a 100,000 of these things happening every two days, right? Like some of them are going to be way out in the and I found I've actually found wallets where um you know the opposite where it claimed like three times in an hour.

What was the what's the change that allows for this much variance in the amount of uh like how quickly these the the swap or claim happens?

Well, it's it's a distribution, right? So, it's it's any you know like any statistical distribution sometimes, you know, the math is you're not excuse me. So the the claims are it just happens mathematically and so if on average it's 32 you know we get most of them within 5 days but you know a certain percentage a very small like but if you're doing 200,000 of these um if there's 200,000 wallets a couple of them are going to you know fall way out in the you know 0.00001% 01% range of 10 days.

So, it it doesn't happen often, but sometimes it does. And sometimes those people notice and they're like, "Why did it doesn't actually hurt you? You're still earning the alpha and you're still earning the tow. It just doesn't get put into your wallet except until 10 days have elapsed.

What is the reason for this variance?" Like it used to be every what I think you said 72 blocks or 360 minutes.

Yeah. 360 Yeah. What was the the change off of that?

So I think the big reason is if you keep so when you keep um let's say you've got you know 10 to staked um and you're going to earn a little bit of dust in subnets one through 128. Um there's a minimum amount of um alpha you can hold, right? You have to hold I think it's 0.00005 or maybe it's 0.00005 alpha before the chain says you can't have that.

So if you're earning this little bit of dust if they did a claim and you couldn't hold it, what would happen, right?

Yeah. You don't get to keep it. So by making it two like if you have 10 tow it might actually take you two days before you have enough where it's actually claimable for you to actually earn it. So by waiting 2 days you can it'll actually claim versus if you tried to do it every you know 72 minutes it just would be ready you know it'd be like the it'd be like when you're you know you're constantly you know are we there yet? Are we there yet? No you're not ready yet. We're not ready yet. you know, and then um so by doing it every two days, then you start getting your rewards and it works.

Okay, that makes sense. It's like it's like the concept of I'm not going to give you a tenth of a penny. It's like you got to wait till you have a whole penny at least.

You got to wait till I have a whole penny, right? I you know, if I start giving you a little cut up pennies, you're like, what am I going to do with this? I don't. So, we're going to wait till you have enough that you can actually claim it.

What do you think is the biggest problem that Bit Sensors yet to solve?

In terms of like how the chain works or like in terms of like they're trying to do or real general opened question and wanted to see where you took that.

I don't know. Um I would love to see what you know I have trouble keeping