“With great power comes great responsibility, especially when it comes to shaping the minds of artificial intelligences.” - Ted Chiang in The Lifecycle of Software Objects

While blockchain is changing the way we transfer value, AI is changing everything, at least according to some, including how we work, interact, date, and gain knowledge. 

Yet, not everyone is in on the hype. Take it from this Australian engineer who uses flowery prose to describe the levels of violence he’ll go to if you mention AI in his vicinity. If a topic inspires those levels of rage, it’s evident that too much has been talked about it, and a significant amount of grifters have appeared. 

A familiar pattern for anyone in crypto. After all, blockchain is often the first hype object the current boom is compared to. And at the beginning of the year we witnessed countless tokens rebranding to AI for a better chance at number going up. 

Moving on from that, blockchain projects now focus on the intersection of AI and blockchain, where using decentralized tech can create real value. 

To explore what’s possible, the SQD team invited Michael Heinrich, Co-Founder and CEO of 0g Labs, and Evgeny Vakhteev, CEO and co-founder of Guru Network, to give their take on the space's current state. 

You can listen to a recording of the discussion here or read on for an overview of AI x Blockchain, and insights from the space. 

Recently, there has been an increase in YouTube videos and blog posts that talk about the limits of AI and that GPT-4 is actually getting worse. Maybe that is just because it’s being trained on what humans input, and statistically, that trends to mediocrity. So the speculation. 

However, our speakers had a more optimistic view of the current hype. Michael explains that it’s justified until we get AGI, but one has to be critical of companies whose sole innovation consists of wrapping GPT API calls with a different interface. He highlighted that AI signifies a change in the computing paradigm. If fully adopted, we would not interact ourselves, but have AI agents do things on our behalf. 

And we might not be limited to getting help with digital tasks either, as he foresees the broader availability of AI-powered robots this decade. 

“Well, Klara (a robot). Since you appear to know so much about it. Will you please reproduce for me Josie’s walk? Will you do that for me? Right now? My daughter’s walk?" - Kazuo Ishiguro, Klara and the Sun 

Referencing the AI hype in crypto, Evgeny added that while big companies spearheaded the trends with their own custom models, we’re also seeing an increase in the availability of devices powerful enough to handle AI. This creates an opportunity to build dApps with personalized and custom user flows. This would remove the excuse that UX is the bottleneck once and for all. 

According to SQD CEO Dmitry, AI will eventually be commoditized and have little moat. Instead, he predicts that whoever can access better-structured data will have a moat. Since data grows faster than any one company can process it in a centralized fashion, it needs to be handled in a decentralized network. This also makes it openly accessible and composable, creating a more equal playing field. 

After all, things are currently working, right?

Except, only a handful of big companies are reaping the benefits from training their AI models on data that creators haven’t even consented to being used. That in itself is ethically questionable enough, yet impossible to prevent as an internet user. 

The next issue with large, widely used models is that they lack transparency. Even experts cannot explain why models produce certain outputs. Bias is somewhat pre-supposed, and so it might enforce certain values of a centralized company instead of presenting the most accurate answers. 

“If we want to use AI in a societal context, be that for manufacturing or admin, you would want to know how the algorithms work. You want to have visibility into what an AI agent is doing and a guarantee that it’s indeed aligned with what’s good for humans.” - Michael, 0g Labs 

Another worry with centralized AI is the way they source data. Take the recent outrage surrounding Adobe, who announced that anything you created in Photoshop might now be used to train their AI. Meta and other Big Tech companies training their own models are not different, and although Instagram claims you can “prevent” them from using your pictures, you’ll never truly know. The one way to fight it is to create a lot of shitty images. After all, garbage in garbage out applies to all cognitive processing.

“People worry about what they can and should share with these AI apps, which prevents some use cases from being realized. We’ve launched a crypto chart app in the Bloomberg Terminal with the same analytics as the Dex Guru Web version. However, we cannot release this feature in Bloomberg due to compliance and fear that data is shared that shouldn’t be. In such cases, data ownership by individuals secured by cryptography could greatly help.” -Evgeny, Guru Network 

Speaking about social contexts, that’s, after all, where most of us waste spend our time online doomscrolling and shitposting. 

And even though some projects have gone on to tokenize tweets, selling them for millions during stupid bull season, people don’t actually care much about owning a specific tweet. 

Instead, Dima points out that what matters more is the feed and transparency over who owns the content I see on my timeline. With an increase in AI-generated content and countless parties able to spread misinformation, it’ll become important to see what type of information has been created where and by whom. 

"What if the powerful can use information abundance to find new ways of stifling you, flipping the ideals of freedom of speech to crush dissent while always leaving enough anonymity to be able to claim deniability?" - Peter Pomerantsev in This is not Propaganda

AI could help curate customized feeds that, for example, only include posts by verified humans. This is only possible with open social graphs like Lens and Farcaster, where all the data is publicly available, and anyone can aggregate via a client only what they want to see. 

There are countless use cases for AI and blockchain, from combining AI’s analytical capabilities with onchain data to authenticating images, videos, etc. As discussed, blockchain can also bring a level of transparency to AI models, especially when the data they are trained on and their code are on-chain. 

Michael believes that once the infrastructure is in place, AI agents can do anything humans do onchain, from trading in DeFi to checking in on your portfolio to minting NFTs. What’s more, AI could enable what they pitched us with intents: you tell the agent you want to go from asset A on chain B to asset X on roll-up Y, and it just figures out how to do that at the specified price without causing you headaches. 

The Guru network team's focus is on UX. They began by building a B2B product suite and then expanded to retail. Already now, AI plays a big role in the product, starting from an AI-generated NFT collection that is used to access and track services to the orchestration and automation layer that facilitates decision-making and makes it easier to trade, at times, across chains. Eventually, Evgeny also believes this will be the key to adoption: showing new ways how AI can be used in the context of blockchain to make life easier. 

SQD CEO Dmitry, on the other hand, dreams of an AI agent that can automate airdrop farming and yields for him, as he’s usually too lazy to do these things. Why aim for mass adoption by humans when bots vastly outnumber them anyway? 

On a more serious note, he shared his excitement for our current research project: client-side indexing which creates a completely new way to consume data locally. If done correctly, it means users could work with their own AI models locally, increasing privacy and independence from third parties. 

AI is here to stay. 

For a full recording of the discussion go here. 

And if you’re looking to access onchain data to train an AI model, contact our team. We’re happy to help. 

“With great power comes great responsibility, especially when it comes to shaping the minds of artificial intelligences.” - Ted Chiang in The Lifecycle of Software Objects

While blockchain is changing the way we transfer value, AI is changing everything, at least according to some, including how we work, interact, date, and gain knowledge. 

Yet, not everyone is in on the hype. Take it from this Australian engineer who uses flowery prose to describe the levels of violence he’ll go to if you mention AI in his vicinity. If a topic inspires those levels of rage, it’s evident that too much has been talked about it, and a significant amount of grifters have appeared. 

A familiar pattern for anyone in crypto. After all, blockchain is often the first hype object the current boom is compared to. And at the beginning of the year we witnessed countless tokens rebranding to AI for a better chance at number going up. 

Moving on from that, blockchain projects now focus on the intersection of AI and blockchain, where using decentralized tech can create real value. 

To explore what’s possible, the SQD team invited Michael Heinrich, Co-Founder and CEO of 0g Labs, and Evgeny Vakhteev, CEO and co-founder of Guru Network, to give their take on the space's current state. 

You can listen to a recording of the discussion here or read on for an overview of AI x Blockchain, and insights from the space. 

Recently, there has been an increase in YouTube videos and blog posts that talk about the limits of AI and that GPT-4 is actually getting worse. Maybe that is just because it’s being trained on what humans input, and statistically, that trends to mediocrity. So the speculation. 

However, our speakers had a more optimistic view of the current hype. Michael explains that it’s justified until we get AGI, but one has to be critical of companies whose sole innovation consists of wrapping GPT API calls with a different interface. He highlighted that AI signifies a change in the computing paradigm. If fully adopted, we would not interact ourselves, but have AI agents do things on our behalf. 

And we might not be limited to getting help with digital tasks either, as he foresees the broader availability of AI-powered robots this decade. 

“Well, Klara (a robot). Since you appear to know so much about it. Will you please reproduce for me Josie’s walk? Will you do that for me? Right now? My daughter’s walk?" - Kazuo Ishiguro, Klara and the Sun 

Referencing the AI hype in crypto, Evgeny added that while big companies spearheaded the trends with their own custom models, we’re also seeing an increase in the availability of devices powerful enough to handle AI. This creates an opportunity to build dApps with personalized and custom user flows. This would remove the excuse that UX is the bottleneck once and for all. 

According to SQD CEO Dmitry, AI will eventually be commoditized and have little moat. Instead, he predicts that whoever can access better-structured data will have a moat. Since data grows faster than any one company can process it in a centralized fashion, it needs to be handled in a decentralized network. This also makes it openly accessible and composable, creating a more equal playing field. 

After all, things are currently working, right?

Except, only a handful of big companies are reaping the benefits from training their AI models on data that creators haven’t even consented to being used. That in itself is ethically questionable enough, yet impossible to prevent as an internet user. 

The next issue with large, widely used models is that they lack transparency. Even experts cannot explain why models produce certain outputs. Bias is somewhat pre-supposed, and so it might enforce certain values of a centralized company instead of presenting the most accurate answers. 

“If we want to use AI in a societal context, be that for manufacturing or admin, you would want to know how the algorithms work. You want to have visibility into what an AI agent is doing and a guarantee that it’s indeed aligned with what’s good for humans.” - Michael, 0g Labs 

Another worry with centralized AI is the way they source data. Take the recent outrage surrounding Adobe, who announced that anything you created in Photoshop might now be used to train their AI. Meta and other Big Tech companies training their own models are not different, and although Instagram claims you can “prevent” them from using your pictures, you’ll never truly know. The one way to fight it is to create a lot of shitty images. After all, garbage in garbage out applies to all cognitive processing.

“People worry about what they can and should share with these AI apps, which prevents some use cases from being realized. We’ve launched a crypto chart app in the Bloomberg Terminal with the same analytics as the Dex Guru Web version. However, we cannot release this feature in Bloomberg due to compliance and fear that data is shared that shouldn’t be. In such cases, data ownership by individuals secured by cryptography could greatly help.” -Evgeny, Guru Network 

Speaking about social contexts, that’s, after all, where most of us waste spend our time online doomscrolling and shitposting. 

And even though some projects have gone on to tokenize tweets, selling them for millions during stupid bull season, people don’t actually care much about owning a specific tweet. 

Instead, Dima points out that what matters more is the feed and transparency over who owns the content I see on my timeline. With an increase in AI-generated content and countless parties able to spread misinformation, it’ll become important to see what type of information has been created where and by whom. 

"What if the powerful can use information abundance to find new ways of stifling you, flipping the ideals of freedom of speech to crush dissent while always leaving enough anonymity to be able to claim deniability?" - Peter Pomerantsev in This is not Propaganda

AI could help curate customized feeds that, for example, only include posts by verified humans. This is only possible with open social graphs like Lens and Farcaster, where all the data is publicly available, and anyone can aggregate via a client only what they want to see. 

There are countless use cases for AI and blockchain, from combining AI’s analytical capabilities with onchain data to authenticating images, videos, etc. As discussed, blockchain can also bring a level of transparency to AI models, especially when the data they are trained on and their code are on-chain. 

Michael believes that once the infrastructure is in place, AI agents can do anything humans do onchain, from trading in DeFi to checking in on your portfolio to minting NFTs. What’s more, AI could enable what they pitched us with intents: you tell the agent you want to go from asset A on chain B to asset X on roll-up Y, and it just figures out how to do that at the specified price without causing you headaches. 

The Guru network team's focus is on UX. They began by building a B2B product suite and then expanded to retail. Already now, AI plays a big role in the product, starting from an AI-generated NFT collection that is used to access and track services to the orchestration and automation layer that facilitates decision-making and makes it easier to trade, at times, across chains. Eventually, Evgeny also believes this will be the key to adoption: showing new ways how AI can be used in the context of blockchain to make life easier. 

SQD CEO Dmitry, on the other hand, dreams of an AI agent that can automate airdrop farming and yields for him, as he’s usually too lazy to do these things. Why aim for mass adoption by humans when bots vastly outnumber them anyway? 

On a more serious note, he shared his excitement for our current research project: client-side indexing which creates a completely new way to consume data locally. If done correctly, it means users could work with their own AI models locally, increasing privacy and independence from third parties. 

AI is here to stay. 

For a full recording of the discussion go here. 

And if you’re looking to access onchain data to train an AI model, contact our team. We’re happy to help.