If you thoroughly read through the blocmates research report for 2024 (authored by Grant – Chief Larp) that was released earlier this year, you would’ve come across the paragraph below from the Crypto and AI section – detailing what Grant had to say about the Crypto and AI narrative:

“I thought this would do well last cycle but mid-curved it to death. I knew it would be an easy, normie-friendly idea, and I still think it is. The problem is I am not smart enough in that area to understand what is vapourware and what isn’t.”

So, how about we fix that for both Grant and the rest of you curious heads? After some caffeine-boosted research, I feel smart enough to unleash what we know about this new, buzzing narrative that has since swamped the crypto space.

If you’re reading this article with so much vim and vigor, I bet you are eager to see what part we mention your bags — nothing wrong with that, if we’re to be honest.

However, as much as it is important to adopt a left-curving mindset when Bitcoin has finally blown through the roof of its 2021 all-time high after a teeth-gnashing bear market, we also think it's equally vital to scrutinize what truly resonates within a narrative versus what is mere hype — especially when it has to do with what Vitalik Buterin describes as “the two (Crypto & AI) main deep (software) technology trends of the past decade.”

Guide through the Vapourware

When examining a narrative that draws heavily from an industry beyond crypto and blockchain, like DePin, RWA, and now AI, it's crucial to initially segment or classify the narrative's interaction with crypto.

To accomplish this, we suggest examining the intersection of Crypto and AI from two viewpoints:

1. Crypto-enhanced AI - Where crypto ‘balances’ out AI tech
2. AI-boosted or assisted crypto - AI solutions to crypto, cryptography, and blockchain operations.

In the first category, the idea is that crypto’s nexus with AI is anchored to the fact that AI systems aren't open source and require intensive data and computing power that bigger centralized companies (Nvidia, Google, etc ) monopolize.

These inherent limitations to centralized AI result in siloed development of models and hinder overall progress, given the barriers posed by huge financial commitments. Moreover, with growing concern about how AI might affect human views positively and negatively, there's a real chance to explore decentralized AI systems to tackle this issue.

On the flip side, one might consider that if we're concerned about the closed nature of AI models, why not make them open-source? Decentralized systems aren't the only way a project can be open source. Despite how plausible this sounds, a counterargument exists that an open-source AI model is susceptible to attacks through simulations that could deceive the model. The prospect of a hacked AI model is disconcerting, especially considering how deeply AI can mimic human behavior. No one wants the events of the iRobot movie to play out in real life.        

(We recommend reading the section “AI as the rules of the game” in Vitalik’s most recent blog post.)

It’s within this line of thought and framework we've recognized that the core value propositions of blockchains may actually improve Al systems. The core tenets of decentralization, permissionlessness, immutability, and transparency help avoid the downsides of extractive centralization, high costs, and essentially an in-real-life iRobot event.

We identified this post below that mentions a few subcategories, some of which are areas where Crypto enhances AI and others where AI boosts crypto:

Crypto-enhanced AI

Crypto used to decentralize access to AI compute (GPUs):  

This category stems from the inefficiencies in the cost and availability of hardware for AI development. As AI projects scramble for compute systems such as GPUs, projects are springing up to cover the gap in the market, offering decentralized compute solutions. Some of the projects that fall in this category are:

• Akash - A supercloud platform that provides permissionless access to cloud resources. Akash allows users (Tenants and Providers) to supply and buy computational resources. Akash operates as a secondary market for the demand in AI compute.

• IONET - A compute cloud network that allows AI/ML builders to cluster hundreds/thousands of GPUs together for intensive workloads. Read about it more in our Solana DePIN report.

• AIOZ - A distributed infrastructure network that allows anyone, anywhere, anytime, access to decentralized services such as storage, AI computation, content distribution, VOD/live streaming, and even IPFS/IPFS pinning. (Find out more in our recent article)

• Aethir Cloud - Building a scalable decentralized cloud infrastructure (DCI) for Gaming and AI. (Alpha: Their decentralized node sale starts on the 20th of March, 2024)

• Render Network - Connecting AI GPU rent seekers and 3D artists with near-unlimited GPUs in a decentralized global network.

• Others include Nosana, Opsec, Netmind, Inferix, iEX_ec, CUDOS…etc.

Crypto as a method for decentralizing machine learning training

This niche takes advantage of crypto decentralization to enhance aspects of artificial intelligence through democratization, decentralization, and incentivization of AI training. One way this is done is by creating a decentralized machine learning services marketplace that connects AI model training requests with those willing to train the model – sort of a P2P approach. An example of a project providing this service is GensynAI.

Gensyn accounts for four participants in the process: The submitters who submit requests for training, the solvers who carry out the actual model training, the verifiers who check if the tasks are done properly by the solvers, and lastly, the whistleblowers who act as watchdogs to keep the verifiers in check.  

You can view Gensyn as some form of AI model training-as-a-service platform that also provides AI computational tools. In this regard, Gensyn is a really interesting project to watch.

Bittensor also provides another impressive approach to decentralizing AI model training. While the project is quite complex to understand, one simple way to look at Bittensor is in how similar it is to the way Bitcoin works — Bittensor is a proof of work (POW) network attempting to be the "Bitcoin of AI," where they train AI models by having a bunch of builders compete for each block to see who answers questions the best while rewarding participants.

How does this work? Essentially, Bittensor developed a decentralized P2P marketplace that allows miners (AI builders) to square off with one another within sector-specific Bittensor subnets. In return, the miners with the best models are rewarded with $TAO using Bittensor’s unique YC consensus mechanism.

The thread below goes into detail about Bittensor's underpinnings.

Crypto used as an incentive to Bootstrap AI compute  

As we’ve mentioned earlier, a gap in the AI industry is looking to be filled with sufficient computing for AI training. A few projects are getting innovative in ensuring that they meet the demand of these AI labs or trainers by incentivizing participants or providers. One such project is Grass, which scrapes websites when installed on users’ PCs or phones. This data is then collected and stored to help train AIs.

Another example is Synesis One, where participants can earn $SNS by completing micro-tasks, incentivizing the contribution of AI training data.

Using Zero-Knowledge Machine language (zkML) to verify workloads

At the most basic level, we have all interacted with AI — mostly in queries to large language models (LLMs) like ChatGPT or X’s Grok. These LLMs feed on data that provides answers or replies to our queries.

However, have you ever wondered about the authenticity of the output of these LLMs? What if they are not as authentic and accurate as they should be? And, given that AI models are trained by human input, isn’t it plausible that a nefarious attempt could be made to tailor AIs to serve specific interests? (*ahem* *ahem* Gemini)

Moreover, if models adopt a closed system to their algorithms, how do we hold them accountable?

Maybe Elon Musk was right all along. We need more transparency with AIs.

This is where the concept of zero-knowledge machine learning (zkML) comes into play — as a verifiable check that holds AI models accountable. As a stand-alone, zero-knowledge is an aspect of cryptography that helps prove computation while hiding the underlying data that you may want to keep secret.

So, the idea for zkML is to create zero-knowledge proofs of these computations behind machine learning models such that we can prove the input and output (end-to-end) of AI models while hiding the model’s proprietary weights. It also works the other way: zkML could be used to obfuscate a user’s query from the model owner - much better than OpenAI keeping a log of every user interaction.

The concept of zkML is very important as it serves as a provable basis for AI’s integration with Crypto, especially in situations involving multiple models.

An example of a project building in this category is Modulus, which uses cryptography to verify that AI results aren’t doctored using specialized ZK provers.

AI-boosted or assisted crypto

This viewpoint is self-explanatory — it relates to all the areas and sectors of crypto and the blockchain where AI improves, as well as new primitives made possible by artificial intelligence, including those building on Bittensor.

Moreso, with the help of provable (zkML) systems, we can say that the boundaries to how AI impacts crypto native products or inherent blockchain services are limitless.

AI as a tool for offering improved DeFi solutions/interaction

Smart contracts lay the foundation for decentralized finance, and by integrating LLMs, they present another way AI makes existing blockchain components more efficient. This is because smart contracts, in their most basic form, are subject to limitations that require constant updates by the developer or a DAO.  However, with the rise of AI models, we are beginning to witness integrations into decentralized financial applications in various forms. Some examples are Mozaic Finance, which uses AI to offer optimized yield to save user time and maximize profitability, and Virtuswap’s Minerva, a proprietary AI-based system for optimizing protocol rewards.

AI smart agents that enhance user experience

There is also a case for developing open-source smart agents that interact with smart contracts to optimize user experience for the end user in various forms, with the user in control by the authority of their private keys. Vitalik, in his recent paper on the intersection between AI and Crypto, also makes a case for the use of smart agents in improving security by verifying smart contract interaction for the user – what he describes as “Info defense.”  One product that catches the eye in this subsection is MorpheusAI, which is designed to incentivize the first peer-to-peer network of personal general-purpose AIs that can execute smart contracts on behalf of a user.

AI-assisted on-chain products

In addition to the methods mentioned earlier, AI boosts crypto through its application in creating superior crypto products and enhancing various on-chain services. For instance, AI can improve data gathering within blockchain networks, conduct on-chain research and analysis, perform on-chain audits, and bolster on-chain privacy measures. Some examples in this category include Aegis, pioneering blockchain security with AI-enabled solutions; ChainML, using AI agents to improve customer service, analytics, code generation, and marketing.

The Spoof

When it comes to emerging narratives in crypto, one certainty is that they will inevitably draw attention from a spectrum of actors, ranging from the commendable to the dubious, with an unfortunate proliferation of the latter.

Similarly, the AI narrative is not exempt from this pattern, as evidenced by the rapid emergence of X accounts that are less than a month old, claiming to offer the next best thing that intersects crypto and AI. These entities swiftly raise funds via platforms like Telegram, adorn their bios with buzzwords synonymous with the AI narrative, and hastily launch tokens—often devoid of tangible products.

However, there’s a pattern to how these pumps and dumps (PnDs) operate, and we will do our best to share what we can to help you stay guarded.

Foremostly, because it is on the blockchain or promises to integrate blockchain components doesn’t mean it can be that cheap to build out– as such, you might want to be wary of any ‘AI’ project that raises an insignificant amount behind closed doors with undisclosed VCs (1000% of the time, they are shady KoLs) and proceed to launch a token. More often than not, you will be dumped on.

Also, from an observatory point of view, one of the most common spoofs in the AI narrative are projects claiming to have built a decentralized LLM. We recommend looking into these projects carefully as LLMs like Chatgpt-like bots are no longer considered ‘special’ or innovative and are probably the most common form of contemporary AI models. Like my blocmate buddy, Dan would say: “If it's an LLM, then it is probably a good time to go grab a Coca-Cola, take a walk, and oooo look…firzbeee”

In addition to what we recommend being skeptical about within the AI narrative, are projects promising to give access to GPUs or other forms of AI computing without any realistic or straightforward direction on how they intend to achieve this uniquely, different from what is already obtainable with projects like IO.NET, Gensyn et.al mentioned in the earlier parts of this article.

Consequently, the trademark of what can be considered the real deal in this narrative is, more often than not, the opposite of the aforementioned probable spoofs — they’ll most likely feature expressive documentation that touches the rudiments of their operations with an explicit conveyance of how they intend to achieve their vision. In addition, you’ll find at the helm a credible team with badges that account for deep-level knowledge of both cross fields. A further look will reveal a significant raise and well-structured tokenomics or incentive model, amidst other things. Most importantly, we recommend looking into what they’ve set out to do vis-a-vis the above-listed intersection points to determine if the project is a well-planned covert operation to harness the buzz or the real deal in every sense.

Moreso, should everything check out, it is advisable to still take your due diligence a step further by keeping an eye on team wallets and vesting schedules to ensure that nothing shady goes on.  

The crux of the matter is that while the narrative may seem enticing for those seeking significant returns on their investments in AI tokens, it's important not to become overtly intrigued by unserious AI projects capitalizing on the narrative. Albeit, we do recommend some left-curving (not financial advice). Ultimately, It would benefit all stakeholders—including you, me, and the entire industry—to prioritize genuine projects focused on decentralizing AI or enhancing crypto through AI tech.

Conclusion

Listen, mate! Not even the blind will fade the potential that AI and crypto present, even at the most basic level. For one, we think it is a very normie-appealing narrative despite how intricate and complex it can get — and going by that line of thought, we do not recommend mid-curving the entire narrative in a market as primed as the one we happen to be in right now.

Nonetheless, within the narrative, we have become fond of some that should significantly impact both industries, such as zkML, GPU DePINs, smart agents, the Near Foundation team,  and Bittensor. The challenge, however, is in the regulatory uncertainty that plagues both industries (AI & Crypto) simultaneously and the tendency to be regarded as a bubble (the latter to which we agree not).

We look forward to seeing how things play out in the long run and what other new use cases meet the demands of the romance between both industries.

[Thanks to 563 for the immense help in bringing this piece to life].