AI Radioroom Episode 1 | Exploring 0G Labs’ Solutions to Challenges in AI
Summary
TLDRIn this AI radio interview, Michael from ZeroG G Labs discusses the vision behind their modular AI chain, aiming to provide ultra-high performance and programmable data availability for decentralized storage networks. With a focus on AI, the project seeks to eliminate barriers between centralized and decentralized systems, ensuring transparency and efficiency, crucial for societal use cases like traffic control and manufacturing systems. The conversation also touches on the potential of on-chain AI, the importance of blockchain for societal trust, and Michael's personal journey from tech and business to co-founding ZeroG G Labs.
Takeaways
- 🚀 Michael from ZeroG G Labs is developing a modular AI chain aimed at high performance and programmable data availability and decentralized storage.
- 🌐 The project is inspired by the need for data throughputs in the tens of gigabytes per second to support onchain AI training, which is significantly higher than the current best-in-class throughputs.
- 🔍 The goal is to remove barriers between centralized and decentralized systems in terms of performance and cost, considering physical limitations like the speed of light.
- 💡 Michael's journey includes a background in deep tech, business, and finance, culminating in the establishment of a company focused on workplace well-being, which scaled to 650 people and $100 million in contract at ARR.
- 🤖 The co-founders of ZeroG G Labs are highly regarded computer scientists with whom Michael connected through a classmate, leading to the establishment of the company about a year ago.
- 🔑 The project's unique selling points are its performance, programmability, and AI focus, with a modular approach that allows for customization of data storage and security.
- 🔄 The technology is designed to be infinitely scalable, with plans to release a second phase of the main net that will allow horizontal scaling of consensus layers.
- 🛡️ Concerns about centralized control of AI models and the potential for misuse or mistraining are driving the need for transparent and verifiable AI systems, which blockchain can provide.
- 🔮 Michael believes that onchain AI is currently underhyped in the broader context but is gaining attention in the web 3 space, with societal use cases being particularly important.
- 🤖 Michael uses AI in his everyday life for tasks like writing answers for interviews, brainstorming, and automating tasks such as coding messages into a CRM system.
- 🌟 The average user would benefit from adopting a decentralized AI model for complete visibility, verifiability, and transparency, especially in use cases that require high trust and accountability.
Q & A
What is the main focus of ZeroG G Labs' project?
-ZeroG G Labs is focused on developing a modular AI chain that provides ultra-high performance, programmable data availability, and decentralized storage network, aiming to bridge the performance and cost barriers between centralized and decentralized systems.
What inspired Michael to start his venture in the AI and blockchain space?
-Michael was inspired by the need for a system that could handle data throughput at tens of gigabytes per second for onchain training and the desire to create an environment where AI models are transparent, verifiable, and controlled by decentralized entities.
What is the significance of the throughput capabilities of ZeroG G Labs' network?
-The high throughput capabilities, tens of gigabytes per second, are crucial for supporting advanced applications like onchain AI training, which requires significantly more data transfer rates than currently offered by most data availability networks.
How does ZeroG G Labs' project address the issue of centralized control in AI?
-By creating a decentralized storage network, ZeroG G Labs allows for open and transparent AI that is not controlled by centralized entities, ensuring that AI models can be monitored and verified for societal use cases.
What is the background of Michael that led him to his current project?
-Michael has a diverse background, having worked as an engineer and technical product manager at Microsoft and SAP Labs, consulted for Fortune 500 companies, worked in portfolio construction at Bridgewater Associates, and built a web 2 company at Stanford.
What was the turning point for Michael to dive into the Web 3 space?
-The turning point was in late 2022 when his classmate Thomas introduced him to the founders of Conflux, who wanted to start a more globally scaled project in Web 3, aligning with Michael's interest in blockchain since 2013.
What are the unique features of ZeroG G Labs' AI chain?
-The AI chain offers high performance, programmability, and an AI focus with modularity, allowing users to determine storage duration, location, data type, and security levels.
How does ZeroG G Labs plan to scale their consensus layers?
-They plan to horizontally scale consensus layers, making it possible to increase throughput by adding more servers, similar to adding an AWS server, which will be implemented in the pH two of the main net by the end of the year.
What are Michael's thoughts on the current hype around onchain AI?
-Michael believes that within Web 3, onchain AI is starting to get overhyped, but from a broader perspective, not enough people in Web 2 are discussing it, indicating that the technology is still underhyped in the larger context.
How does Michael envision the future use of onchain AI in everyday life?
-He sees onchain AI being used in various everyday scenarios, such as personal AI agents for tasks like writing responses, brainstorming, and automating processes like coding messages into a CRM system for better follow-ups.
What are the benefits for an average user to adopt a decentralized AI model over existing solutions?
-Decentralized AI models offer complete visibility, verifiability, and transparency, which is crucial for users who need to know exactly what data and context their AI is using, especially for sensitive applications like educational tools for children.
Outlines
🚀 Introduction to ZeroG G Labs and AI Chain Vision
Michael from ZeroG G Labs joins the AI radio room to discuss his project's inspiration and vision. The company is developing a modular AI chain focused on high-performance, programmable data availability, and decentralized storage networks. The goal is to enable on-chain training for blockchain by providing the necessary throughput, which is currently lacking in existing data availability networks. Michael shares his background in deep tech, business, and his journey to co-founding ZeroG after a series of experiences in the tech industry and a desire to contribute to the web 3 space.
🌐 Modular AI Chain: Performance, Programmability, and AI Focus
The conversation delves into the unique aspects of ZeroG's project, emphasizing performance with the ability to scale to tens of gigabytes per second, programmability that allows users to customize data storage solutions, and an AI focus that aims to develop on-chain model training. Michael explains the technical barriers and the company's plan to overcome them, highlighting the importance of efficiency and the potential for blockchain to provide transparency and verifiability in AI systems.
🔮 The Intersection of AI and Web 3: Potential and Overhype
Michael discusses what drew him to the intersection of AI and web 3, expressing concern over centralized control of AI models and the potential for misuse or runaway models. He believes blockchain is ideal for creating transparent AI systems, especially for societal use cases. The discussion also touches on whether the buzz around onchain AI is overhyped, with Michael suggesting it's overhyped in web 3 but under-discussed in web 2, and he envisions a future where different types of agents are used based on use cases.
🤖 Everyday AI Use Cases and the Benefits of Decentralized AI
The final part of the script explores Michael's personal use of AI in his daily life, including using AI to write responses, brainstorm ideas, and automate tasks. He also addresses why an average user might prefer a decentralized AI model over existing solutions, emphasizing the importance of transparency, verifiability, and the ability to understand the context and inputs of AI models, especially for critical applications.
Mindmap
Keywords
💡AI Radio
💡ZeroG G, Labs
💡Modular AI Chain
💡Data Availability Networks
💡Throughput
💡Onchain Training
💡Decentralized Storage Network
💡Programmable Data Storage
💡Blockchain
💡Web 3
💡Decentralized AI
💡Onchain AI
💡Transparency
💡Verifiability
💡Provenance
Highlights
Michael from ZeroG G Labs joins the AI radio to discuss the future of data availability and decentralized storage networks.
ZeroG aims to provide ultra-high performance programmable data availability and decentralized storage, targeting throughputs of tens of gigabytes per second.
The vision is to remove performance and cost barriers between centralized and decentralized systems, enabling on-chain AI training.
Michael's background includes engineering and technical product management roles at Microsoft and SAP Labs.
He has also worked in business consulting for Fortune 500 companies in technology and finance.
Michael's experience at Bridgewater Associates involved managing a $60 billion portfolio construction division.
After graduate school at Stanford, Michael founded a company focused on workplace well-being, which scaled to 650 people and raised $125 million.
The ZeroG project was initiated after Michael's classmate Thomas introduced him to Conflux and the opportunity to work on a global scale project in web 3.
Michael believes in the importance of open, transparent AI for societal use cases to prevent misuse and ensure ethical governance.
ZeroG's modular AI chain is designed to be infinitely scalable and programmable, allowing users to customize data storage and security.
The project's focus on AI and blockchain aims to create an environment where AI models are verifiable and transparent.
Michael discusses the potential of on-chain AI for societal use cases, such as traffic control and manufacturing systems.
He sees the current buzz around on-chain AI as overhyped in web 3 but under-discussed in the broader context of web 2.
Michael shares his personal use of AI for writing, brainstorming, and automating tasks like coding messages into a CRM system.
The average user may want to adopt decentralized AI for complete visibility, verifiability, and transparency in AI models.
Michael emphasizes the importance of knowing the context in which AI models operate, especially for sensitive applications.
The interview concludes with a discussion of the practical applications of ZeroG's technology and its potential impact on the future of AI.
Transcripts
[Music]
hello everybody Welcome to our AI radio
room today we have Michael from zerog G
Labs joining us how you hey everyone
doing pretty well despite landing at 2:
a.m and you know being jetl I feel
pretty energized just being here so
thanks for wor yeah absolutely
absolutely so um maybe just to start you
want to tell us a little bit about your
project uh your journey what inspire you
to um start your Venture in the space
absolutely so um zuro G is a modu AI
chain the first module AI chain our
first product is uh ultra
performant um programmable data
availability and decentralized storage
Network and essentially the the future
that we saw was that right now a lot of
data availability networks have a
throughput of like Best in Class
anywhere from 1 to 10 megabytes per
second but if you truly want to bring
things like onchain training um to the
blockchain we need tens of gigabytes per
second and so we're many orders of
magnitudes of and so we really from a
first principes perspective thought
through how do we architect A system
that gets us to that throughput so that
really there's no barriers between
centralized and decentralized systems
from a performance and cost perspective
besides the physical limitations like
speed of light and so on so that's in a
nutshell what we stand for um our
mission is to make AI a public good and
how I got to this journey is I started
uh you know in deep Tech I um was an
engineer and a technical product manager
at Microsoft and sap labs and then moved
over more to the business side I worked
for ban company for a couple years where
I consulted Fortune 500 companies in
technology and finance afterwards moved
to the east coast and uh worked for
Bridgewater Associates and the portfolio
Construction Division uh quite fun so
about 60 bill the trades on a daily
basis so not a not a small amount quite
quite a bit there and then finally
decided to go back to graduate school at
uh Stanford and built my first web 2
company there and at that point I was
quite fed up with uh just wellbeing in
the workplace and so built a whole
company around that scaled out to about
650 people 100 million in contract at AR
raised about 125 million for it um so
became a top YC and unicorn company and
then late 2022 my classmate uh Thomas
who then became my co-founder later
basically said hey I invested in this
company called conflux um M and fun two
of the co-founders want to start
something more global scale they're the
best uh engineers and computer
scientists I've ever backed like would
you like to get together and he knew
that I wanted to do something in web 3
ever since investing uh into Bitcoin in
2013 and in lots of icos in 2016 and so
I was like yes let's get together let's
check this out um six month of
co-founder dating later I was came to
the same conclusion like Ming and fun
are the best computer scientists I've
ever worked with and uh basically said
we have to start something and so
started the company uh almost a year ago
at this point and um basically said well
what should we build and so we thought
about well where do we have domain
expertise in what's a major unlock for
the space and what are we passionate
about and so after speaking to quite a
few people in the space we basically
came to the conclusion that I just
mentioned around building a module AI
chain because we wanted to see that
future actually happening where we have
open transparent uh artificial
intelligent that's there for human good
yep and efficient efficiency is key
efficiency absolutely I mean in order
for that to happen you need the data
infrastructure to be in place otherwise
like what's the point um when we started
storing a gigabyte on ethereum like $60
million and of course it's cheaper if
you do it on an L2 and so on but um
doesn't solve the problem yeah okay
that's a good start uh so now we'll get
into like the quickfire questions about
like building an AI sure okay so just
quick Snappy short
answers uh you know like Instagram Tik
Tok kind of
style so takeways exactly right so the
first one is could you tell us more
about the project you're currently
working on and what makes it unique in
the AI have three yeah uh three things
uh performance programmability kind of
AI focus and so that whole modularity
piece is in there um you want me to go
deeper or sure sure uh performance I
mentioned already instead of 1 to 10
megabytes is tens of gigabytes and in
fact infinitely scalable once we release
pH two of the main net end of the year
because we also figured out a way to
horizontally scale consensus layers so
just like flipping on a switch getting
another AWS server we can do the same
thing on the consensus side
programmability because it's based
around a decentralized storage Network
um users of the system can determine how
long to store the data where to store
the data what type of data type to use
how much security they want in the data
so that's uh quite a big differentiator
and then the AI Focus we this is our
first product and then we'll start
adding more and more features so that
eventually we can do go to things like
onchain model training as well many
technical barriers to be overcome there
but um sort of that's the key plan good
stuff all right next
one what Drew you to work at the
intersection of AI and web 3 and why do
you believe this combination has such
potential what what I saw that I was
quite worried about is that if there's a
future where all of the I models are
controlled by centralized entities
what's to stop those centralized
entities to a want to you know do
something nefarious and B if they mistra
these models that these models
themselves uh kind of run out of control
so why can't we create an environment
where we know everything about them uh
where the data came from how it was
labeled where the data is stored what
version of the model is what weights
they have how they're making decisions
and so blockchain environments were
ideal is an ideal situation for that and
I think especially important for
societal use cases
so traffic controls systems
Manufacturing Systems administrative
systems I would want to know um that my
model wasn't trained on how to build a
bomb for example yeah so like specific
tailored training almost yeah specific
tailored training um knowing exactly
having input from multiple community
members around how the governance should
work uh what are the economic incentives
for the model putting it in a situation
where it can't cheat like if a model
tells me hey I picked up all the trash
in this environment and here are my
centralized database records and then
actually none of that happened then
obviously the model is cheating or the
agent that's kind of running on that
model and so on a blockchain environment
something has to happen for there to be
a record and so that can't be faked or
changed lovely yeah three more uh in
your opinion is the current Buzz around
onchain AI overhyped or underhyped and
what do you see as the future of this
technology yes so I'd say in web 3 it
seems at the point where it's starting
to get a bit overhyped but then if you
took take it from a broader perspective
in web 2 not enough people are talking
about it so it really depends on the on
the context and so the future is that
for different type of use cases in my
mind we can use different types of
agents and so if
you're booking a restaurant booking a
hotel you can have a you know agent on
your Edge device that's maybe connected
to a blockchain somewhere but can make
decisions for you you know very small
locus of control not a lot of negative
impact that can happen from that but
then again societal use cases
fundamentally need to be on the
blockchain sure so in keeping with use
cases uh what have you found to be your
favorite ways to use AI just in your
everyday
life well I tried to um have a AI
basically agent I used Cloud 3 to write
a lot of answers to my uh one of my
interviews and um it did a decent job I
still have to scrap about 60% but so I
like to use it that way um and then I
also just for some brainstorming like
for example what what does a good
onboarding document look like now I can
spend 10 minutes come up with a bunch of
stuff or I can just spend 10 seconds get
an initial answer and then say like okay
maybe this is missing this is missing um
so that's another use case and then for
automations as well one of the projects
I want to work on is take a lot of the
telegram messages I have and automatic
Ally kind of code them and put them into
CRM system so it's easier to kind of
follow up and so on brilliant use cases
yeah finally final question why would
the average user want to adopt a
decentralized mod model over existing
Solutions uh web 2 nlms sort of thing
yeah if you want complete visibility um
it goes to this whole you know
verifiability provenance um
transparency if you want to know exactly
what into your model and you have
specific use cases that require you to
know that exactly let's say uh the llm
is toing your child for example and
you'd want to know that the context that
it's toing your child in is a very
benevolent one then I think truly open
ai decentralized ai is the solution
there I would agree that's that's that's
one good solution to be fair um okay so
thank you very much for joining us
Michael um big pleasure hopefully we see
you again soon maybe in Brussels
potentially we will be there good stuff
thank you very much awesome big pleasure
Weitere ähnliche Videos ansehen
5.0 / 5 (0 votes)