Verifiable Compute for AI, ML and More: Web3's New Frontier | Alison Haire at SmartCon 2023

Summary

Alie, CEO and co-founder of Lilypad, discusses the project's focus on verifiable, trustless, decentralized compute as the new frontier for Web 3.0. Lilypad, incubated at Protocol Labs, aims to provide a scalable, trustless compute network with a global marketplace for GPUs, enabling mainstream adoption of Web 3 applications. Alie highlights the practical and ethical challenges of centralized AI and ML, emphasizing the importance of decentralized compute for innovation, accessibility, and data ownership. The project also envisions incentivizing node operators with LP tokens and plans for multichain compatibility.

Takeaways

• 🌐 Alie, the co-founder and CEO of Lilypad, discusses the project's incubation at Protocol Labs, highlighting the company's involvement in the decentralized web 3 space.
• 📈 Lilypad aims to address the current gap in verifiable, decentralized compute, which Alie sees as a critical component for the modern internet and a frontier for web 3 development.
• 🤖 The project focuses on AI and ML compute, aiming to provide a trustless and decentralized platform that can be utilized at an internet scale, facilitating mainstream web 3 applications.
• 🔐 Verifiable compute is presented as a solution to ensure the integrity and security of computations, which is particularly important as AI and ML become more integrated into web services.
• 💡 Lilypad's vision includes creating a new data economy by leveraging idle processing power and providing an accessible, permissionless compute network with a global marketplace for GPUs.
• 🌱 The project is designed to be highly native to running nodes, incentivizing users to contribute their spare GPU resources and be rewarded for their participation.
• 🚀 Lilypad has recently released version 2, with plans to expand to multi-chain compatibility and an incentivized testnet, indicating active development and a roadmap for future growth.
• 💼 Alie emphasizes the practical and ethical challenges of centralized AI, such as ownership, access, and the potential biases introduced by relying on models from a few dominant companies.
• 🌐 The project aims to democratize AI by providing open-source models, which are said to be faster, more customizable, private, and capable, thus fostering innovation and reducing biases.
• 💼 Lilypad's compute network is positioned as an alternative to centralized cloud providers, offering a more distributed and efficient model that could be more accessible and better for pricing and supply-demand dynamics.

Q & A

What is Lilypad and what does it aim to do?

Lilypad is a project being incubated at Protocol Labs, aiming to create a verifiable, trustless, decentralized compute platform. It focuses on enabling mainstream web 3 applications to utilize internet-scale data processing without relying on centralized services like AWS.

Why is verifiable compute considered a core need for the modern internet?

Verifiable compute is one of the three components of the modern internet, but it has been a missing piece in the decentralized internet story. It's now seen as essential for bringing the next billion users to web 3 and for allowing users to do more with their data in a decentralized manner.

What are the challenges in building decentralized and verifiable compute?

The challenges include the complexity of building such a system and the fact that the research is still cutting-edge and catching up. Additionally, the crypto space has been focused on solving more pressing problems like account abstraction.

How does Lilypad plan to make decentralized compute accessible and affordable?

Lilypad aims to create a global, trustless, permissionless compute network with an independent global marketplace of GPUs. This marketplace is designed to eliminate middlemen, keep prices low, and make the service accessible to everyone on the same terms.

What is the significance of being able to run AI and ML models on a decentralized network like Lilypad?

Running AI and ML models on Lilypad allows for ownership of AI models, access to models, and the ability to train models with security and speed. It also supports innovation through open-source projects and addresses practical and ethical challenges arising with mainstream AI adoption.

How does Lilypad address the issue of AI model ownership and access?

Lilypad promotes decentralized AI through distributed networks, which helps prevent central entities from controlling access to AI advancements. It also encourages open-source models that are more customizable, private, and capable.

What are some practical use cases for Lilypad's decentralized compute network?

Lilypad can provide cheaper, more open, better distributed, and more efficient compute power. It can also help with security, data ownership, and provide a stake in the ownership of AI solutions.

What is the 'water lily' project mentioned in the script?

The 'water lily' project is a proof of concept that uses Lilypad to train an AI model on artist input data. Users can then pay a small amount to get images generated by the trained model, demonstrating the practical application of Lilypad's compute capabilities.

What is Lilypad's roadmap for the future?

Lilypad has recently released version 2 with updated models and plans to go multichain, being EVM compatible. They also intend to release an incentivized testnet and aim for a mainnet launch in the following year.

How are participants in Lilypad's network incentivized?

On the supply side, node operators are compensated with LP tokens. The network tracks all participants for future launches, implying that there will be incentives for their contributions.

Is Lilypad's compute network limited to GPUs, or does it also include CPUs?

While Lilypad primarily focuses on GPUs due to the AI and ML focus, it does imply the potential for a network that could include various types of hardware, including CPUs.

Outlines

00:00

🌐 Introduction to Lilipad and Decentralized Compute

Alie, the co-founder and CEO of Lilipad, introduces herself and the project, which is being incubated at Protocol Labs. She reflects on her first crypto conference and sets the stage for discussing Lilipad's focus on verifiable, trustless, decentralized compute. This technology is positioned as a core component of the modern internet, currently missing from the decentralized web narrative. Alie highlights the complexity of building such a system and the fact that the research is cutting-edge. She also touches on the readiness of the space for this technology, mentioning developments like account abstraction and ZK proofs. The goal is to make decentralized compute accessible and part of the web 3 ecosystem, aiming to bring the next billion users to web 3 by leveraging this technology beyond just user experience improvements.

05:02

🚀 Decentralized AI and the Need for Lilipad

Alie delves into the importance of decentralized AI and the role Lilipad aims to play in the space. She discusses the ideological drivers behind decentralized AI, emphasizing the need for open access to AI advancements to prevent central entities from controlling AI models and their usage. The talk highlights the benefits of open-source models, which are more customizable, private, and capable. Alie also addresses the practical challenges of AI innovation, such as the high cost and limited access to GPUs, which Lilipad aims to solve through its decentralized network. The paragraph also touches on the geographical concentration of cloud provider data centers and the need for a more distributed model. The vision is to provide a practical solution to these challenges while upholding the ideological values of web 3, such as transparency, data ownership, and peer-to-peer networks.

10:03

🛠 Lilipad's Roadmap and Future Vision

Alie outlines Lilipad's recent achievements, such as the release of Lilipad V2, and future plans, including multichain compatibility and an incentivized testnet. She discusses the project's vision to create a global, trustless, permissionless compute network with a marketplace of GPUs. The goal is to make this network accessible to everyone on the same terms, leveraging idle processing power to unlock new data economies. Alie also mentions a proof-of-concept project called 'Water Lily,' which demonstrates Lilipad's capabilities in generating images from artist input data. The roadmap includes further development and expansion, with a focus on making Lilipad a key player in the decentralized compute space, offering solutions for AI and ML tasks, and integrating with blockchain technologies for secure and efficient computation.

Keywords

💡Lilypad

Lilypad is a project being incubated at Protocol Labs, aiming to create a verifiable, trustless, decentralized compute platform. It is central to the video's theme as it represents the frontier for Web 3, focusing on enabling mainstream adoption through verifiable compute capabilities. The script mentions Lilypad's role in building a decentralized compute network that can handle AI and ML tasks, emphasizing its importance in the evolving landscape of web technologies.

💡Verifiable Compute

Verifiable compute refers to the ability to ensure that computations, especially those performed on decentralized networks, are accurate and trustworthy. In the context of the video, verifiable compute is a core component of the modern internet and is integral to Lilypad's mission to provide a decentralized compute solution that can be trusted and verified, thus enhancing the reliability and transparency of computations in Web 3 applications.

💡Decentralized Compute

Decentralized compute involves distributing computational tasks across a network of computers rather than relying on a centralized server or cloud service. The video discusses the importance of decentralized compute in the context of Lilypad, which seeks to leverage this approach to provide a more accessible, secure, and efficient alternative to traditional cloud computing services for AI and ML tasks.

💡Web 3

Web 3, or Web 3.0, is a term used to describe the next generation of the internet, characterized by decentralized applications, blockchain technology, and greater user control over data. The video emphasizes the role of Lilypad in advancing Web 3 by providing a decentralized compute platform that can support the growing needs of AI and ML applications, thus contributing to the broader vision of a more open and user-centric internet.

💡AI and ML

AI (Artificial Intelligence) and ML (Machine Learning) are fields of computer science that focus on creating systems capable of learning from and making decisions based on data. In the video, AI and ML are highlighted as key areas where Lilypad's decentralized compute platform can make a significant impact, by providing the necessary computational resources and capabilities to train and deploy AI models in a decentralized manner.

💡Protocol Labs

Protocol Labs is a research, development, and deployment institution that focuses on decentralized technologies, including the creation of Filecoin and IPFS. The video mentions Protocol Labs as the incubator for Lilypad, indicating the project's alignment with the lab's mission to advance decentralized technologies and solutions.

💡Decentralized Storage

Decentralized storage refers to the practice of storing data across a distributed network of nodes, as opposed to a centralized server. The script mentions decentralized storage in relation to Filecoin, another project from Protocol Labs, and how it complements Lilypad's focus on decentralized compute, suggesting a broader ecosystem of decentralized technologies.

💡Account Abstraction

Account abstraction in the context of blockchain technology refers to the concept of separating account addresses from transaction logic, making it easier to manage and interact with blockchain accounts. The video briefly touches on account abstraction as one of the pressing problems in the space that is being addressed, indicating the ongoing evolution and improvement of blockchain infrastructure.

💡ZK Proofs

ZK proofs, or zero-knowledge proofs, are cryptographic methods that allow one party to prove to another that they know a certain piece of information without revealing the information itself. The video mentions ZK proofs as an emerging technology that is catching up and becoming more integrated into the space, suggesting its potential role in enhancing privacy and security in decentralized compute environments like Lilypad.

💡Data Economies

Data economies refer to the systems and marketplaces where data is bought, sold, and exchanged. The video discusses Lilypad's vision to unlock new data economies by leveraging its decentralized compute platform, allowing for more efficient and innovative ways to manage and monetize data, particularly in the context of AI and ML applications.

💡Open Source

Open source refers to a type of software or project where the source code is made available to the public, allowing for collaborative development and modification. The video emphasizes the importance of open source in the context of AI and ML, suggesting that open source models are more customizable, private, and capable, which aligns with Lilypad's goal to democratize access to AI advancements.

Highlights

Alie, co-founder and CEO of Lilipad, discusses the project's incubation at Protocol Labs.

Lilypad aims to create verifiable, trustless, decentralized compute for Web 3 applications.

Decentralized compute is identified as a missing piece in the decentralized internet ecosystem.

The complexity of building verifiable decentralized compute and the readiness of the space are highlighted.

Decentralized storage through Filecoin is now available, setting the stage for verifiable compute.

Lilypad's vision is to build an internet-scale distributed compute network for data processing.

The project offers a global, trustless, permissionless compute network with an independent marketplace of GPUs.

Lilypad's approach to decentralization includes ideological and practical use cases.

Ownership of AI models and access to hardware for model training are key challenges addressed by decentralized compute.

Decentralized AI through networks like Lilypad can prevent central entities from controlling AI advancements.

Open source models are presented as faster, more customizable, and more private alternatives to centralized AI.

Access to GPUs at a reasonable price is a significant barrier to AI innovation that decentralized networks can help overcome.

Lilypad's two-sided marketplace aims to improve efficiency and pricing for GPU access.

Decentralized physical infrastructure networks can provide a more distributed model than current cloud providers.

Lilypad's practical applications include cheaper, more open, and more efficient compute power for developers.

The project also focuses on transparency, data ownership, and community ownership of AI solutions.

Lilypad's proof of concept project, Water Lily, demonstrates the potential of decentralized AI for content creation.

The roadmap includes releasing Lilypad V2, going multichain, and launching an incentivized testnet.

Participants in the Lilypad network are incentivized with LP tokens for supplying GPU resources.

Lilypad is EVM compatible, allowing for direct smart contract interactions and results computation on-chain.

Transcripts

[Music]

uh so hi everyone uh I'm Alie uh I'm

from uh lilipad uh co-founder and CEO

this is a project that's being incubated

at protocol Labs though uh which is why

you might have seen me at filecoin and

protocol Labs events um so we're

currently uh incubating and building

this out at protocol Labs as well uh

really great to be here actually fun

fact uh back in 2020 I think chain link

conference 2020 was one of my first ever

crypto conferences that I went to it was

Co so it's was all online and as an

Australian I could actually attend so I

got to meet like some of the gang back

then and now I'm actually talking here

so uh pretty cool uh anyway moving on uh

so what is lilypad what is this

verifiable trustless decentralized

compute there's a lot of big words in

there why do I think it's the now

Frontier for web 3 and why am I shoving

a i and ml onto another slide we've

probably had enough of hearing those

words but there's actually a reason for

this and hopefully uh you'll see that by

the end of this chat um so verifiable

compute is kind of a core need it's one

of the three components of the modern

internet uh but at the moment uh it's

kind of Compu has kind of been a missing

piece of the decentralized kind of

Internet story um so there's a few with

a few experimental um kind of exceptions

along the way and that's likely for two

reasons one is complex building

decentralized compute that's verifiable

is complex uh and the research is only

is kind of still bleeding edge but

catching up now secondly I think as a

space uh we just weren't quite ready yet

and there was more pressing problems to

solve like some of the things that we've

seen this morning um you know a account

abstraction a big one I'd like to be

able to use uh or or send this compute

or Ai and ml compute model that we're

building to um you know everyone we want

mainstream adoption um so we're starting

to see the results of some of these

discussions in the space like come out

like the account abstraction Pieces come

out uh ZK uh proofs have come out now

all that sort of thing is finally

catching up of course uh I've worked for

filecoin so uh decentralized compute uh

sorry decentralized storage is also now

available as well um so I think now

we're ready we're ready for verifiable

compute it's the now Frontier of web 3

and it's going to enable us to actually

we've heard a lot about bringing the

next billion users to web 3 and the ux

story is part of that but what about

being able to do stuff with our data

like I said it's one of the core pillars

of the internet uh so we want to be able

to do that where it doesn't actually

connect to just another AWS service um

so that's what I'm aiming we're aiming

to do here at lilipad we're trying to

create this project uh of verifiable

trustless computation uh like this

internet scale and can allow for these

mainstream web 3 applications to use it

uh so um we want to unlock this uh new

data economies as well and as

kryptonians I didn't know is that a word

I just made it up um we also are highly

native to running our own nodes as well

so why not be rewarded for your spare

GPU being used by a project like this as

well um so we're hoping this is a space

that the Lily W pad Network can play a

big part of um so our vision is to build

this internet scale uh kind of

distributed compute Network that enables

internet scale data processing so not

just Ai and ml but those are big parts

of it especially with the mainstream

adoption we're seeing and other

arbitrary computation from blockchains

while unleashing idle processing power

and unlocking this new wave of data

economies um so in Practical terms

lilipad offers kind of three main things

a global trustless permissionless

compute network with an independent

Global Marketplace of gpus so we're

basically running a two-sided

Marketplace which we want to do in

crypto we want to get rid of the

middleman we want prices to stay low we

want things to be accessible which is

the next point we want it to be

accessible to everyone on the same terms

uh so it's permissionless and thirdly we

want to use this distributed Hardware

network with inbuilt cryptographic trust

mechanism so offchain large scale

compute with onchain guarantees which is

you know probably a lot of parallels

with chain link here which is

great uh so why not AWS then why aren't

we using that why do we need this um

probably don't need to tell you all you

can probably guess already but it's not

just a buzzword uh like with AI and ml

reaching the kind of mainstream adoption

curve uh first with kind of generative

AI images and now with llms uh like chat

GPT um there's several kind of practical

as well as ethical challenges that are

arising here so some of these are

uniquely suited to a decentralized

compute solution like lilypad and these

kind of include ownerships of a

ownership sorry of AI models access to

AI models and Hardware to train the

models and security and speed to

Innovation through open- Source projects

so I'll expand a little on some of those

firstly the capability for decentralized

AI with distributed uh networks like

lilipad is one of the major ideological

wi's and it's a driver for me personally

so we've all heard this story before

being in this web 3 Community uh Bears

repeating here though in the same way

that one person should not control

access to social media or one Co company

shouldn't control access to your data or

scientific breakthroughs Central

entities probably should not be allowed

to control access to AI advancements or

then Lobby for how those should be used

um along the way so currently both the

models and our access to them are either

a black box or rate limited or there's

kind of other questions surrounding that

uh so that's one reason that networks

like these are going to be important in

the future and what's also interesting

to add to that point though is I don't

know if anyone's read this article um so

this was like a leaked Google article um

about how they have no moat and neither

does open Ai and basically what it says

is that open source models are faster

they're more customizable they're more

private and they're more capable um so

that means opening up access to AI gives

a better Pathway to Innovation it also

reduces biases because we aren't just

like relying on um an algorithm created

by one of these companies which are

generally American companies no offense

to Americans but you know what I mean um

and you know uh we're opening up access

instead uh for open Innovation and that

means also that people can get there

quicker because they know what their use

case is they know what they need one of

the big problems here though for being

able to innovate in the AI space is

access to gpus is at a reasonable price

I don't know if anyone's tried to use a

GPU lately on one of the main

centralized Services they're really hard

to get to and they're really expensive

anyway um so deep in or decentralized

physical infrastructure networks uh and

one of those is vcoin uh can help solve

this with kind of an open compute

Network so leveraging these networks and

connecting them to a legitimate uh uh

front end and code protocol base really

uh can help us uh solve this kind of

practical problem of access to

gpus um this should given it's a

two-sided Marketplace also be good for

efficiency and pricing so Alig this

Supply and aligning this supply and

demand uh by providing these gpus and

new revenue streams to the other side of

that as well so olops don't make for

proficient use of hardware and they

don't provide good pricing models for

users um this is another great point I

don't know if if anyone knows Brooklyn's

alinker or the fishing team um but she's

amazing CTO there uh she also made the

point that if you look at these large

Cloud providers their data centers are

really in two main places almost

entirely and there's not data centers

anywhere else so I think uh this this

kind of distribution model needs to be

looked at as well uh so a quick repap on

some of my points uh hopefully I'm not

all ideology here there's actually some

real practical use case reasons which I

think the team before was very are

adamant about having a real use case I

think there's some real use case reasons

uh for this decentralized compute

Network as well as some uh kind of

ideological uh solutions that I think

are very important as well um so we've

probably heard a lot about Ai and

blockchain recently this you know web 3

is really good at unlocking Network

effects eliminating these middleman and

creating uh peer-to-peer networks so

this is what we can bring to web 3 and

web 2 developers Main stream developers

so cheaper more open better distributed

and more efficient compute power

combined with kind of this transparency

and data

ownership and having a stake in the

ownership of AI uh Solutions as well

there's also a security point there

which I think don't know if anyone was

at Allison uh jman's I'm sorry if I

pronounced that wrong uh talk this

morning but uh yeah it was really good

right she was talking all about uh

security and the need for more people to

be able to look at this as well uh so

these are some of the things that we can

do uh that a web 3 is uniquely capable

of helping AI with uh and there's

definitely AI problems cropping up um so

we can create open source Community

ownership here uh we you know we can

create proof of humanity in Providence

and I think we're already seeing

projects doing that as well uh and

there's also attribution models so uh

crypto as a payments layer is incredibly

efficient uh so we can use it for things

like giving attribution to the original

content creators and this is a proof of

concept uh project we did uh with Lily

Pad hence the name water lily which

basically takes uh artists input data uh

trains it so does a Laura training model

on it and then a user can go ahead type

in their prompt pay a small amount of um

money and get their uh images back this

is on an old uh stable diffusion model

that's why the images don't look as good

as the new ones do uh which you'll be

able to see uh you can try it out

actually it's all live um I don't think

I've got any pictures in this slide

though um anyway so our road map uh

we've just released lilypad V2 which has

a lot of updated models uh we intend to

go multichain as well we're evm

compatible so anything evm we can run on

um we're going to release an

incentivized test net as well uh for

suppli so if anyone's interested in that

please hit me up uh and May net launch

next year okay ah there we go um we just

did a whole team did a talk on

everything we're we're uh doing at Lily

Pad as well so if you want to know more

uh you can have a look at Phil Dev

Summit uh and uh have a look at those

talks sorry that's an old picture it was

meant to be Sal Salvador uh Dar picture

of Barcelona which is really kind of

weird um but this one's from Iceland so

hence the

Aurora okay thank you everyone uh hope

open to questions if anyone has any

otherwise try it out one question sorry

no questions I'll take one

though uh yes there's a whole uh

research paper on that so maybe I'll

just give you the link to that because

I'm getting like

the thank you hello sorry quick question

if you have a look in our docks there's

oh there's a quick question up there too

hey yeah so uh how the participants are

incentivized do do you have kind of

you're talking that you used to

compensate those who are in this

calculation like Network yeah yeah

exactly so on the supply side uh the

node side they are compensated with LP

tokens currently we're only at test net

though so those are pretty worthless but

we are tracking uh all participants on

the network for the future launch oh

cool there network of GPU and also I

guess CPU like mostly mostly gpus at the

moment um and yeah mostly gpus that

we're using because we're focusing most

of our modules or jobs are around Ai and

ml at the moment so Laura training or

infant training um yeah and when you say

that it's EV compatible you mean that

you have also some contracts and the the

tasks that uh we have from our contract

we can uh put it through the contract

directly on chain you can directly run

from a smart cont get results calculated

then also like yep you can get yep you

can run this in a yes or from a solidity

contract the Oracle part is provided by

Chen link yeah I mean no it's not quite

an oracle uh I'm kind of hoping in

future we might be able to use ccip as a

messenger between our multi chains

actually um that's probably one of the

ways I'd integrate um but we're

currently yeah there's there's no chain

link cool great thank you I'm asking

because in the next session of what we

are doing we will need scoring okay that

uh has to be calculated off chain in ml

models and then

recorded into this would be completely

useful for that and it would be highly

compatible with being put with either ZK

models like to do the computation or

with oracles um to help do computation

so yeah great use case thank

[Music]

you