Tom Trowbridge on decentralized computing and the shift to cloudless networks

0:01I'm ashon Addison from the cryptocoin show today on blockchain interviews we have Tom Trowbridge CEO of fluence Tom welcome to the show and thanks for taking the time great thank you for having me as and happy to be here you're very welcome excited to dive into uh llms blockchain decentralization not always on chain which I think is an interesting perspective that many also

0:21forget about um and what fluence is working on and and the leadup to the main net as well I would love to start off our conversation by first of all hearing a little bit about yourself your background leading up to fluence what got you so interested in this then we'll dive into everything fluence certainly so um before fluence I help found a um L1 called to D hash

0:40craft uh and I joined Manson Leman pretty much when it was a white paper um and was you know a key part of building the team raising the money getting the first council members and helped kind of get everything going from the white paper to the launch and that was a you know intensive multi-year kind of baptism by fire into the whole kind of blockchain

1:07decentralized you know world and ecosystem so it was a terrific learning experience it gave me um also it also was a great educ technical education because both Dr L Barett and man Harmon um were incredibly um credentialed and impressive technologists and and frankly scientists and so I learned a terrific amount about the whole industry through that lens which was interesting um and

1:35then when I left um you know hadera there was really important lessons I gained from that and you know the technical fundamentals being solid is number one and so that was kind of key point the other was not over promising and kind of delivering what you said you were going to deliver especially in this industry I think is really important and some people kind of take that for

1:58granted but I don't think you can hear here and the other and this is something I think I wanted to improve on from hadera experience was make sure you are very crypto native and very dialed into that Community because it makes an enormous difference in terms of the time and attention that people pay on your play you know pay on your project and so fluence had those three characteristics

2:25and a lot more um but those were those were kind of some some kind of experiences and um and lessons I took from that wow very cool and I do love hyera I'm actually working with some projects on the Hyer ecosystem still so it's really great to hear that you you were part of uh the the conception of it um so that's very cool and now moving into fluence how exactly is that

2:49different and what exactly is fluence working on in the industry right now um well fluence we call fluence cloudless Computing Network and so you know if you take a step back we think that we realize that um Computing started off in people's homes and people's offices and then over a 20 year period people were brought onto the cloud and companies and

3:12people have moved to the cloud you're using your Gmail that's happening in the cloud you're you know Netflix stuff is delivered from the cloud like everything has moved a lot not everything but a lot has moved to the cloud and that has been a cost savings and it's probably been a security enhancement given the profiles of Hosting most of these um applications

3:32and and and servers inhouse but that has come at a cost a a economic cost a centralization cost a censorship cost and a resilience cost and so you now are in a world of a several companies that basically own the cloud and so we think and they're also terrifically profitable companies obviously as well and they provide a valuable service so that's not

3:56not um you'd expect nothing else but the internet itself the economics lead to Central and so because benefits of scale are rewarded and so we think we've gone from the in-house servers to the cloud but the next and final iteration is cloud list and what that means is a network that is open and isn't constrained to any one provider where your application lives

4:23wherever somebody will host it and uses whatever open source um or frankly closed Source um comp components that's anyone will host and if if the host that's on has an outage or goes down it can go somewhere else if you decide you want to leave that host you can go leave that house immediately you don't need to unwind and you don't need to worry about

4:44being kicked off for a violation of an 80 page term of service that is arbitrarily interpreted differently one day for a particular reason so that that's the iteration and fluen is building is this cloudless um Computing Network work and stack and that is what we're um been building for years and are very close to launching wow very cool and yeah I feel like so many people use

5:10the cloud right now and they just they don't really know how it works or where their files are who's actually controlling them they might know yeah it's somebody else's server computer but um I feel like it's they don't know how good it could be because the infrastructure hasn't really got there yet you know there's I feel like eventually there will be the extin

5:32extension of the cloud as we know it where you know there's just that handful of companies that controlling it and it will be a decentralized cloud and and probably people will be able to capture some of those rewards that are so centralized right now well I mean listen there's there's a whole bunch of elements that you mentioned there that I agree with and we should spend a minute

5:52of time on I mean first people use the the best technology you don't know and care how it works because it just works right you and I don't know the details of the iPhone but we're happy it works the way it does cloud you don't need to know how it all works but the problem is that the larger companies get the more intertwined they become with governments

6:12and this goes back to the railroads I mean it's not any company's fault it's just happens that that is the case and so you end up being you end up having kind of a couple things happen but you have censorship potential and that's one thing but you also the bigger you get the fewer other options there are you end up with sort of oligopolistic pricing yeah which is

6:31sort of similar to monopis pricing we have very high margins you see these companies massive margins and customers are you know use it they're happy they're but they're they're locked in and if you want to leave it's a little complicated and the other point is that they have prevented the open um an open architecture that would drive more Innovation and so you even look at

6:55application at the application layer where you had compan companies like Twitter or Facebook that used to have open Integrations and those are now were cut off and gone and so we see a world where people build applications that are open and where Integrations are easy and cannot be disabled and so I think there's a a a whole generation of open applications which will build on an open

7:22Cloud list stack that leads to kind of an innovation acceleration and so on the one hand the you the cloudless stack helps because it saves cost it reduces lock in it has all these other benefits it also has verifiability because you're using blockchain it's not onchain but because blockchain is part of the stack you can validate and verify the computation was performed which you

7:47can't do in a cloud stack you a company can certify it and that's useful but it's very different than actually having a cryptographic proof that something happened so there's benefits to to the kind of infrastructure layer if you will and then there's benefits to the application layer of where you can have open applications that share um data and users in a way that can't be disabled or

8:08cut off and I am a believer that that is going to lead to a terrific amount of innovation that we we sort of have a difficult time imagine now it's very exciting and thank you for those uh insights Tom now with the new version of like the cloudless more decentralized Computing Network I've I've seen Solutions pop up but because of the network effect of things

8:33like you know Gmail or or Google Drive and these storages people are so familiar with that and also they've refined the the user interface you know for Grandma and everyone's able to use it uh I feel like you have to build a product in in web 3 that's very you know as you said you don't need to know the inner workings in the back um and it also needs to be very easy and and

8:55userfriendly to adopt and of course it's one thing for somebody who's hosting an application they probably have more technical expertise to host their app or website but when you want to do personal file storage or things that you know people have no technical background how can they how can you lower the barriers to entry on a cloudless decentralized

9:14Computing Network to allow for people to adopt it quickly and transition from those Cloud networks that have all those issues but people are sort of stuck because they don't want to change because they don't want to learn anything new I think we have to separate two elements which is the infrastructure element from the application element and so we are primarily targeting um the

9:36infrastructure layer which is in your example would be the developers of the products that would compete with like say the Google applications you mentioned all right so but what we're doing is our objective is to make the building of those applications so simple and to commoditize the backends to to such a degree that instead of 50 to 60% of developer resources spent building

10:07backends 20 perent is or 10 percent is so you can have you know 80% focused on the front end on the user experience on the user interface and so we won't be building and and and working on the applications you're referring to but our view is that the the simplification for developers that we will enable will facilitate more

10:35investment in the front end and will also drive more people into the ecosystem to compete to build better products so that that that's that's that's where our role in that is no that's that's a great point and you know I'm not a developer myself but I've worked on applications I've used the Amazon web services back end and it is ridiculous how you know you you almost

11:00need to take a four-year degree to figure out how many functions they have and all these different functions and and they suck you in that it's like I I I spent years trying to close my account but there's 100 different things and it's very complex um I feel like it could definitely be simplified and also it could definitely be cheaper um if we had more decentralized

11:19networks yeah I mean it will definitely be cheaper I mean to give you sense um and and here's another kind of um misconception it's important to clear up and people think of decentralized networks or they think of you know decentralized storage or compute sometimes the default is to think of people using their laptops or their phones we are running in kind of tier

11:39one data centers where there's excess compute capacity so you know we toured a Data Center in Las Vegas um about four months ago where Netflix has hosted you know a whole number of of of well-known companies are hosted and so I think that's kind of one misconception that I want to clear is that this decentralization does not sacrifice the institutional quality and resilience

12:07that I think is required these days so that is um that's that's kind of an an important important Point yeah know that's that's it's great to know because especially with the the quality of the apps and the amount of traffic that's coming in you're going to need things that are going to be able to compete with the warehouse level of of uh computing power well exactly sorry that

12:29was my other thing I forgot the train of thought which was also price and so you just have to look and see what the um margins are of the cloud providers to see the incredible opportunity there and so when you are have racks that are unused in the data center your cost your marginal cost is pretty much the electricity you're using and maybe some depreciation as well right so there is a

12:54terrific amount of margin potential that goes to the customer or when you're able to use these and the key thing is about fluence is that if you went to a Data Center and said I want these you know thousand servers and I want them for two years you pay you know a reasonable amount to rent them but if you said you can give them to me for a week at a time

13:20you will pay a dramatically lower price for a week to week contract and the reason why it doesn't matter for fluence is because it's an open network if they decide to go offline the applications move automatically to a different provider and so what that does is it commoditized that compute because it can go anywhere which leaves those providers with no pricing power and also allows

13:46them to monetize at the same time they can monetize in small amounts their excess capacity without signing up for a long-term contract at a price that they may find some customer willing to pay more than so you have two benefits that an open protocol has a sort of compelling rationale for why the pricing is significantly lower than a centralized you know provider that has

14:12to invest in those servers buy them hold them for a long period of time or else rent them from somebody else no that's a really great Point Tom because that's how part of the way that they have in traditionally sort of sucked you in in that you know lock lock people in for longer term contracts give them a bit of a discount on pricing um and you know

14:33other smaller Alternatives would cost more and then you're sort of stuck and then once you've built your application there the migration process of something different is like it's almost not worth your time you you even if New Alternatives come out that are cheaper in the future you're sort of stuck with your application because you don't want to pay to migrate and and or learn new

14:50technology yeah definitely so now with the Computing Network do I have it right that it's there's blockchain involved for but it's not fully on the blockchain sort of there's web 2 use cases and web two technology as well as web 3 yeah so the way flu fluence is a offchain um compute Network and so the way to think about that is to take is take a step back and realize that dist

15:17the decentralization way prated the blockchain and you know mentioned that um Napster was the first decentralized application at scale and this may surprise people but you know what do you think napster's users were at Peak you know um Millions 80 80 million 80 million right and so here we're we're we're you know clawing for thousands or tens of thousands of users or a million users

15:44for certain blockchains here in Napster 10 years before Bitcoin had 80 million users a peak so that shows product Market fit for decentralization and so that is you know obviously was it was had other some material issues but the um the point is that our our um solution is offchain because there is 99% of the web is non-deterministic compute which means it

16:11does not it will not work via smart the smart contract so smart contract compute is fully decentralized for sure on ethereum no argument there and that has terrific use cases for primarily Financial applications great but tons of other applications like you're not going to have have an Uber figure out the right route for you to take via Smart contract it's just not going to work and

16:35so you need to have non- determinist to compute so fluence compute is offchain like file coin storage is offchain but it's validated verified reward and incented onchain and so we use blockchain where you know to maximize its strengths we don't figing blockchain and force a use case into it no I think that's a good uh way of going about it because sometimes you know at at the

17:02current state of where blockchain is it's it's more efficient obviously and there's a reason that you guys are just doing the verifying and and and part of it on chain but you know there's a reason that the internet has been working so well for the last 20 years without blockchain as well there's still a lot of good things about it not just internet I mean internet for sure but

17:20just computation is not all it cannot all be done in a deterministic manner and that is I think the most important piece of it and so you know you kind of impli you know what what use cases are out there and so the other important thing about fluence is it's CPUs which are central processing units which are you know again 98% or whatever it is of the web when

17:42people talk about AI they they're everyone talks about gpus and you'll see a lot of GPU networks spring up of people that want to train um artificial intelligence AI llm learning language models and that is a huge piece and I'm we're very excited about that and the technology we're building on fluence is relevant for that um in the next phase but we're focused on CPUs and CPUs are

18:06relevant in two ways to AI one is that you need to clean the data before it gets into the model and to clean the data you need CPUs that's more the generalized compute it's not the GPU side and then flute is also relevant because you want to be able to prove that your model was trained on particular data set and you've seen the New York Times Sue openai claiming

18:33copyright infringement because they claim that open a trained on New York Times maybe did maybe didn't I don't know but I can promise you that if you're building an LM now you're be worried about copyright infringement and lawsuit risk and so you're going to be able to provably validate the data or provably demonstrate what your model was trained on and that fluence is is

18:55capable of that as well and the last piece is um then once you have this model I think it's important that models run in a decentralized way so they're less subject to manipulation and to opaque manipulation and so you've already seen um chat GPT you know start to have you know let's just say questionable answers you know can carbon give me five reasons why carbon is good

19:24for the human race you know you may not get the answer you want you know there's there's a whole number of those what will it say about I don't know Pakistan versus India you know I don't know but they're centralized companies are going to come under a lot of pressure to generate specific answers or forc answers in models depending on where they are and that is going to

19:46universally undermine credibility and so a decentralized model will certainly have its failures but if it's trained in an open transparent way verifiably trained on certain data then you and the answers are provably um um generated from that data set you at least can't claim it was manipulated and I think that will lead to those models to be the most trusted models eventually and I

20:14give the obvious example here is is people say well you know open ey has you know 10 billion from Microsoft and it's an enormous company how could anyone compete with it but you think about Linux that had I don't know Microsoft had a 10 year Head Start I'm not even sure and pretty much unlimited resources and Linux just through the you know just through the open the global developer

20:37Community has become the default standard for operating systems and so I think we have a model there to show what open source can do the power of it and I think that's going to be applicable here with but it also has to be run on these open systems in order to help prevent manipulation down the road once it's complete yeah no that's a great and very

21:00important Point Tom because people are starting to use chat gbt more and more every day and you have to not just rely on the answer it gives as the absolute truth you know I feel like from the first version of chat that came out you could ask the same question to it now and the answer is probably different whether they've updated their algorithm with new information or just new bias um

21:23it's hard to tell where the data sets are and and what's the bias you know whether it's about climate change countries political things um i% yeah so I feel like decentralized Computing networks are are definitely going to play a role in that um and I mean actually and just that point as they become viewed more and more used they become inevitably viewed as somewhat of

21:47a source of Truth the way Google does except instead of Google giving you links it's giving you an answer right so that is going to make governments in particular super focused on it you ask a question for example imagine during Co asking should I wear a mask you're telling me that the government isn't going to care what that answer is right like and when when that

22:09right so that's just one example you know whether marry mask makes sense or not just as an example of the type of um very specific question for which there aren't links there's not data it's just an answer and I think you will see that more more and more focused on that going forward yeah no it's a it's a great point so now with more of a decentralized llm or open- Source chat

22:32GPT competitors how would does that work where they go to fluence for example as an open network they could host the application and the data sets public or how would that work couple different ways so first um the data um First Data could be clean influence they come they come influenc they ask us or they don't ask us but they go on the um to our protocol and

22:56you know upload a lot of data and find and BAS basally find hosts that are willing to um for a certain whatever price it is clean that data and make it consistent um then they would train that data but that data training on a model would be GPU based but not not on fluence at least not in this iteration of fluence fluence will be good for training will be GPU let's say

23:18compatible in version two which is probably about a year is away I'm not sure exactly but everything we're building is GPU compatible um we just started this before gpus were where is relevant um and then they can also then come back and say okay we have this model can you validate it was trained on the correct data and prove that and then ask host who will host this data and

23:43then then run um and then pay to run queries on this data and so it's basically the running of the model it's the computation the quering of that model that host on fluence are suitable to do and that is and that is the important part and this is another Nuance to what decentralization means decentralization doesn't mean I have

24:10five providers and I pick one and there's some people in the ecosystem that do that like I can do that with my web browser what decentralization means is I pick or the the the the algorithm the network or me pick some provider but if that provider goes down the application automatically fails over to another provider without having to reprovision it and so decentralization

24:37means effectively self-healing it means resilience and that's the important piece here because if you just have to go and keep repr provisioning it's less usable it's less user friendly and it's um you know they're barriers to the um to to using and adopting it well I'm very looking forward to trying some of those uh competitors and you know it's even chbt is so new and it's growing so

25:06quickly I feel like the decentralized competitors and hopefully more competition in general will be able to be spun up quickly and compete you know headed head-to-head with uh some of the more centralized llm systems yeah I mean I I think that's I think it's going to happen um you know it's uh it's all all in progress right now MH so so now about fluence

25:30specifically you mentioned there about version two I want to roll that back to sort of the present day and what's coming up in the near future the coming months here uh I understand that the main net launch is coming soon can you talk about where exactly you're at and what is the main net launch mean and what will happen at that point yeah mainnet launch we're scheduling for eth

25:48Denver on February 27th so we're going to do a you know IRL and real life event um at eth Denver very excited about it and that means that we'll have a network usable stable enough for people to use it also means and this is relevant that we will launch the fluence token and the fluence token is critical because we have to pay people to provide their

26:13compute to the network and so just like filecoin offers fill rewards for the storage providers SPS they call them we provide the FLT reward to providers who offer their computes the network and then as more and more um useful compute enters the network that useful compute pays you know Fiat I mean in stablecoin

26:41but you know denominating Fiat prices to the providers who then take that Revenue as opposed to the FLT reward and so our view here is we want to make sure that providers are compensated fairly um for their their services and have downside protection so we're pricing this in a in a Fiat way as well but also

27:09benefit from the upside as this as it um as the use of the network grows and as um and as we hope the um the amount of locked fluence increases significantly as well wow very exciting that's right around the corner what is the best way for people to follow along with with those updates if they can't make it to eth Denver specifically uh is there online community or the updates and and

27:32how we can follow along with the main net launch yeah I mean listen I think first I would certainly go you know our website is um is fluence.com to um to stay stay in touch

27:59with us amazing uh I I'll put those links in the show notes below appreciate your insights into all of computing and and blockchain Tom it's a very interesting discussion um looking forward to the main net launch and we should definitely follow up in the near future after that great well thank you so much really appreciate it and look forward to uh look forward to coming

28:17back thank you again Ashton