Welcome back to a new AMA with Arbitrum.
Today we have a very exciting AMA with Lil.ai and Churro is our guest house and Arbitrum
contributor that will be leaving this space for today.
So yeah, let's get started and thank you so much for joining us.
GM, yeah, thank you for the intro Ana and really excited about this AMA, especially
because this is focused on AI and community management.
So it's such an interesting mix because I haven't seen AI play a role within community
So I'm really excited about this AMA and to get started, I want to start off with some
So Cassin, do you want to give us an intro about yourself?
Yeah, thank you so much for having us.
It's amazing to be here with our Arbonauts.
My name is Cassin, I am the CTO over at Lil.ai.
The project itself has been running for just under a year and it's really been a self-built
All of the co-founders and the team come from blockchain and web free, either in community
management of the profession or various technology backgrounds and our backgrounds and
We had a very interesting year and I'm very excited to be able to be here today to really
share what we're planning on doing this year and as you say, roll out our community
managers using machine learning and AI.
Thank you very much for having us.
Yeah, any time, our pleasure.
And yeah, I saw you guys have been like doing really cool stuff on that front.
So do you want to talk a bit about Lil.ai, like what inspired you guys to build out that
product and yeah, maybe give us the high level of what can users expect?
Really, that's why I've been in blockchain since probably about 2017 time.
So I've seen a few ups and downs, worked with various projects on various things.
And the thing we really noticed was it was very difficult for projects to expand, roll out,
build awareness, build trust, keep communities engaged and up to date for people.
We saw thousands of people being subject to, you know, victim to scams, misinformation,
because there's small projects, independent small bodies, not particularly verified like a big
And it's very easy to trust some of the things you see.
And we really weren't happy with that.
And while everyone was focusing on maybe more conceptual new technology, we really saw this
huge hole and we couldn't ignore it.
It was such a crucial space, especially when we want crypto to be widely adopted.
And if we can't protect people at a social level whilst enabling projects to grow and focus
on building, the space just won't take off.
So our aim has always been to automate as much as possible, protect as much as possible
by removing scams and spams and just ensuring that for both projects and people, they're
able to have this like single source of trust that they know when they're getting into
something or they're giving out information that it's maintained.
To give you an idea, we first rolled out our first product back in probably mid last year.
And it was a simple, simple idea.
Could we use machine learning to improve on anti-scam and anti-spam?
And we've processed around, I think around 400,000 messages, give or take, and we've
deleted over 7,000 spam messages.
It's 2% of all messages we've seen have been scams and we've been able to delete those
and keep communities on topic, safe and clean.
Yeah, it's really interesting because I think back in 2021, like at the height of the
bull market, right, you've seen tons of discord communities getting hacked out of
And I think one of the reasons that is is because of the bots they rely on to help secure
And yeah, it was such a crazy time, too, because not not only you've seen that in
discord, but you also saw it across Twitter.
And I wanted to ask a follow up question if the little AI is meant for moderating
discord or telegram or Twitter accounts, is it meant to help secure those three
fronts or are you guys primarily focused on one of them?
So the simple answer is all of the above.
We started building out on telegram.
That was where most of our current clients were based and people that we worked with
So our kind of primary role that was telegram.
However, we're currently building integration for discord.
And the idea will be that your human moderator won't need to exist.
You'll be able to rely purely on our tools to act as almost the human, you know, for
passerby, they won't really know they're speaking to an AI is the idea.
In terms of Twitch and Twitter and YouTube and all those things, they are all services
we're also looking to integrate.
Twitter is an interesting one because you can't moderate comments and things in the same
way you can maybe if you're an admin on a Reddit server or hosting your own Twitch
But we are definitely able to automate a lot of those processes, flag issues.
For example, if there is a scam on Twitter, we can automatically flag it and send it to a
project team so they can then try and act on it with Twitter.
But so we're trying to roll out as widely as we can, even outside of web free.
I mentioned Twitch, for example, there's so much opportunity to assist artists and
developers and any community manager to just improve the experience of their
Yeah, it's really cool to see that how you guys are focused on those fronts, because that's
where mainly all the scams take place.
So it's really cool to see how AI is playing an interesting role, especially with machine
And I want to ask about the process of building little AI, were there any challenges you
guys faced along the way?
And how did you guys solve those issues?
Well, I can tell you there are challenges every day.
Machine learning in particular only knows what you give it.
So we're always seeing a slight nuance and a message or a slight tweak, something we've not
So every day we're looking at messages, analyzing all of the chats in our store from our
That's the best way to say it's it's all about constant adaptation.
In terms of building, we've actually found it very pleasant.
Most of those platforms have fantastic APIs.
The hardest issue or challenge, I guess, was working on what tools to use.
There are so many tools and language models that you can use.
I won't go into detail about the ones that we've ended up using and building, but we've
we've really had to work hard to make sure that every tool we've picked is robust and
I think you cut off a bit.
Yeah, it really came down to just making sure those tools were robust enough for what we
wanted, which obviously over time new tools will come out.
We'll have to adapt again.
We'll constantly be pushing ourselves to make sure that our product is as good as it can
be. And we're very fortunate to have such great feedback from our customers.
We're currently running a closed beta, for example, on our AI manager, soon to be released
end of Q1 and all of those community manager, human community managers are pushing it for
its paces, testing it, asking it weird questions, trying to break it and break it.
And I guess we're very fortunate to have that kind of direct customer engagement.
Yeah. And how's that been so far?
What are some feedback that you guys received from the community managers that are testing
it? It's been very good so far.
It's again, like I say about kind of the machine learning side, there's always ways to ask
questions and ways that they expect answers.
If you think there's a good story on open AI and they basically put loads of data in and
found it wasn't what they wanted.
And they had to spend a lot of time fine tuning that model to give the kind of the
performance that it has today.
We've generally had really good feedback, though.
People are asking when it's going to be deployed.
They're begging us for discourse.
So we're working on that as fast as we can.
And we've had fantastic testimonials.
We had one project kind of briefly had a story on called PropBase, which is an RWA out of
Southeast Asia. They were launching their token back in November and they noticed a lot
of kind of fraudulent activity.
And they reached out to us.
We deployed into the channel straight away.
And I think we we processed something like three or four thousand messages in an hour.
We deleted around 300 fake contract addresses and fake pools of people trying to derail this
project. All in an hour, we didn't even know it was happening.
We just were sitting there one day and a flood of deletes came into our system.
And we enabled that project to keep their data lines clean.
And they've been an incredible supporter ever since.
And we can't wait to roll out our community manager to them as well and help them really get to the
Yeah, that's super cool and awesome of like how fast they acted in removing those messages,
because I'd imagine like if you have a huge community, you probably have about four to five
And I'd assume that probably would take so long to just like go through each channel,
especially in Discord and figure out which messages are not the scammy ones and like how
deep within their Discord those scams links are.
So, yeah, I think it's awesome that little AI was able to jump in there and help really
fast. And that's cool to see.
Hmm. Yeah. And it's like you say, you don't just have one moderator, particularly for small
projects. They need to have 24 seven coverage, ideally in multi-language and need all of those
moderators to be up to date with the information they have.
So you then normally need to manage the moderation group, all of which costs money and takes time
away from doing what they're meant to be doing, which is building their platform.
So with those, they can upload all their data they want, they can control the tone of voice
they want, and they really can let our product look after their community for them.
Like I say, 24 seven uptime fully multilingual, built in various built in tools.
And then anything that does require human intervention, we can escalate automatically or
forward those messages on to some of the more back end services.
It's really, for us, just been such a success.
And so we're so proud to be able to support as many people as we can.
Yeah, and I think one thing that people would be worried about is like, what if the AI goes rogue?
Is that even possible? Just start sending out random messages in the chat.
Not so much. We were always monitoring it for a start. And the the moderators in the group always
have the power. So unlike of a bot, where you deploy that bot, and it works in the channel,
and all the commands go through the channel, and the bot fills that channel with spam.
Let's be honest. We use a separate kind of more back end service, which shows you all of those
actions. Every every project can immediately change the tone, stop the bot entirely if they had to.
But what we, the way we've approached it is that we only act on messages that kind of require it.
So our bot isn't constantly talking. It's assessing whether someone needs help, or whether some
information is legit, whether links are appropriate, whether, for example, if you're trying to didn't
support price talk, you might want to just gently direct people to the right kinds of channels.
But we very much focused on a responsive approach to ensure we're just giving people
information as needed, and otherwise being silent.
Yeah, I was just thinking about the worst case scenario. So that's why I bought it up,
because AI is new. And you know, you see AI across different verticals, not just in
community management, but also in popular areas like music, art. So yeah, it's really interesting
how far we've come with AI. And I also know that you guys are focused on other areas as well
with little AI, such as data, Oracle, and as well as Bluebird, which generates an instant
Twitter text. So do you want to dive into the other products you guys are working on as well?
Absolutely. So we released our roadmap earlier this month, and that's available on our website.
If anyone wants to go take a look, we're unleashing a whole kind of raft of amazing services.
So in Q1, we'll be releasing our community manager, as well as jumping into Discord for
the first time, which is going to be great. In Q2, we'll be looking at really building
key technology integrations. So looking at ways to bring in external data sources,
look at unchanged transactions and enable kind of automated payment type services.
And then obviously upgrading all of our systems as we go. A key thing that we'll be releasing
in, I think it's going to be Q3, is these idea of sentiment signals, which is quite interesting.
If you think we're in every chat with every social media platform that that chat is linked to,
we have access to every message, all the member trends, as well as other activity. And we're
actually able to assess the sentiment in that community at any given time. And that data,
anonymized, of course, will be available as its own service, if that's something people want to use.
We're then looking at, as you say, data oracles. Data is so crucial, particularly the right data
and trustworthy data. So we've got some amazing friends over at Parsec, Quick Intel, Databot.
We've got some awesome, awesome partners so far. And together, we're able to just have this
incredible data bank that we're able to use. And that will be available to the business side of
community management and the consumer facing side as well if it's required.
Those sentiment signals I mentioned in Q4 will likely do a commercial release.
So you may end up seeing those available as an indicator, maybe on exchanges or at least some
of those kind of more charting tools that are around. And then we've got a few other projects
which I can't really talk about. But it's definitely branching slightly out of the
kind of community management front, really building LIL.AI out as a big data provider,
which is very, very cool.
And a DeFi project that I want to try out, right? And I want to be able to see how their protocol
is doing based off of on-chain activity. It'd be cool to have an AI that could signal that,
oh, their product is doing good. You should definitely check this out based off of social
activity as well as on-chain. That'd be pretty cool to see. And obviously, there's a lot of
data that tracks that. Like DeFi Llama, you could see the TVL and the users.
So it'd be cool to have that aggregated in one place.
Yeah, absolutely. The big challenge with all this data is what do you do with it?
And I think for us, we're lucky to have so many options for ways we can deliver that data,
ways people can interact with that data, and then building on that data ourselves to try
and build more tools and more services.
Yeah. And as well as the user experience, I feel like it would be a lot more better
because obviously DeFi Llama is heavy data-oriented. So you have tons of data that is
all over the place. But yeah, it would be cool to have that in one place where you could just
go to that place, and then you'll be able to see the projects and the sentiment indicator
that shows if the project is doing well or not. So yeah, just to get out there.
Yeah. And I think the interesting thing about looking at community is it responds to a different
speed than anything else. You get that hype from the devs speaking to the community, teasing things,
and the community will respond. The vibe in that community, good or bad, will respond faster than
anything else. So rather than looking at different indicators and things, this is a complete other
level, particularly when crypto is so community-focused. It's all about people. It's all about
small groups of communities and holders. And we really think there's so much value
on top of, obviously, the management of those communities and making those projects prosper
is just visualizing that and making that available in itself.
For sure. Yeah. I'm super excited to see how this all comes about in terms of seeing other projects
use little AI for their community management and other social media data gathering. So I think
that's gonna be cool to see from you guys. And on the topic of AI in general, what are you most
excited about outside of what you guys are building? Ooh, that's a good question. I did
a spaces a few months ago, and similar question came up. And I think it's where it's most exciting
is where you don't see it. There'll be so many opportunities for, you know, general things in
life to just get a bit easier. And the world is a pretty tough place right now, I think, for a lot
of people. And I think where I'm really excited to see is just efficiency, bringing down costs,
making things faster, making, you know, getting rid of that noise. When you search for something
on Google, you just get so many options, it would be nice if it knew what you wanted and just gave
you what was suited to you. And then I think we're seeing a lot of AI coming into DeFi now,
which is amazing. Blockchain, as we know, compared to kind of traditional finance technology systems
is so much faster, so much cheaper. And I think having those tools to make it accessible to
particularly to new people, and people who are a bit afraid, let's say, of some of the kind of
do-it-yourself nature of crypto right now, making that more normalized and more accessible is,
I think, really going to shape the space. I'm very excited to see
how that pans out, particularly in the DeFi space.
Yeah, same here. And I'm also excited to see how AI plays a big role in gaming,
because something I've seen within the Arbitrum ecosystem, there's a game in particular,
just like Sims, it's called Land Labor, I forgot the other, but it's called LLC, oh,
Land Labor and Capital. And they're a game on-chain game on Arbitrum Nova, which is a gaming
chain on Arbitrum. And they're using AI for generating in-game assets, like graphics,
for the game. So that's something really cool to see. And I think the cool part about that is
that in-game asset AI that they're using, it evolves over time. So it's cool seeing the aspect
in terms of how AI is playing a role in gaming. So yeah, I'm really excited to see how AI plays
a role in both gaming, DeFi, as well as community management too. So super hyped about that.
Yeah, that sounds really fun. I think games over the years have become somewhat a little bit
monotonous. So in order to have a dynamic aspect, I can imagine being incredibly engaging.
Yeah, yeah, really, really bullish. Yeah, and yeah, and I know that we're towards the end
of the AMA. So I want to give you like an opportunity to mention anything else
anything else that you want to say to the community, like where can they go to learn
more about little AI? Do you guys have a Discord, Telegram? Yeah, sure. So I guess firstly,
thank you to everyone who's tuned in. Thank you to Arbitrum for hosting. It's been lovely
to be able to speak to everyone and give everyone an update on what we're doing.
Anyone that wants to check out Lil.ai, we've got a Telegram channel currently,
and a Twitter, obviously, well an X now. And pop into the Telegram channel. We will be on Discord,
of course, in the next few months. But we're always online, we're always loving to chat,
we have a fantastic community on Telegram. And we can maybe send people snippets of our
beta bot and give you guys a flavor of what you'd be able to expect, as well as see some of that
data. So we're able to show you in real time, kind of some of the things we're up to. Our website,
of course, has a roadmap, which everyone's more than welcome to take a look at. And please get
in touch. We want to hear about everything everyone wants to do. So we can help improve
and enhance that experience with people. Awesome. Yeah, really excited about everything
that you guys are building here in this space. So yeah, wish you guys tons of luck. And yeah,
let's go. Awesome. Well, I'm sure we'll speak to each other very soon.
Yeah, for sure. All right. Yeah, have a great Wednesday, everybody.
Have a good week, people. Thank you very much.