Empire Newsletter: How crypto can solve AI’s centralization problem

Why AI needs crypto

Crypto is very good at one thing: getting humans to do one thing. This will become incredibly important in an AI world.

Technology is deflationary and creates abundance, but it is also a centralizing force. Look at Google ( owns 90% of search ) and social (TikTok, Meta).

Chris Dixon documents this phenomenon well in his latest book “Read Write Own.” Unlike internet protocols (TCP/IP, HTTPS, SMTP), internet service companies like Google and Meta have amassed monopoly-like market shares. These companies own your data and, after achieving scale, adopt monopoly-like practices to squash competition. Meta shut down their App Store, OpenAI did the same, and Google has crushed multiple calendar-app startups.

In theory, you can use other social media platforms, but none achieve the same level of network effects as TikTok or Meta.

Read more: AI has a monopoly on power and privacy. Blockchain fixes this.

The centralizing effect of technology may be even greater with AI . I recently came across Anthropic’s CEO interview on CNBC where he estimates “we’re going to see models trained in the next year are going to be about $1 billion…and then 2025, 2026, we’re going to go to $5 billion or $10 billion. And I think there’s a chance it may go beyond that to $100 billion.” Few companies will afford that.

Perhaps more importantly, companies that own the data will have an edge to train models. A startup may be able to raise a couple billion, but they won’t be able to easily replicate clinical trial data or years of user behavior data.

The promise of technology is that it is democratizing and lowers the barriers for companies to reap the benefits. This narrative rings true with AI — after all, companies will be able to embed LLMs and other forms of AI (robots, automation processes etc) to unlock productivity gains. Like trickle-down economics, trickle-down tech will benefit many. But the centralizing effect of technology will likely be no different and perhaps even greater with AI.

So where does crypto fit into all of this?

Crypto has proven to be very good at getting humans to do a specific action which benefits the overall network. From Bitcoin to DePIN, crypto networks are case studies on how to scale networks more efficiently by transferring the costs to do-it-yourself (DIY) participants incentivized to spend resources (time, hardware, compute) expecting to earn in return tokens which represent ownership in the network.

The challenge is that bootstrapping and scaling the supply side of networks can be very capex-intensive in these networks. Historically, this has been one of the reasons why there is network centralization. You have 1-2 ride-sharing companies and social media platforms. But when you radically lower the financial burden to bootstrap a network from a company/protocol to participants, then you level the playing field.

You can more easily overcome the cold-start problem networks face by attracting supply and demand because both sides are incentivized to earn tokens (i.e., ownership in the network).

Giving network participants a greater cut of revenue/profits will be the reason why crypto networks from DePIN to AI will emerge as leaders and will be able to displace incumbents.

One area I see this already playing out is in training — specifically, reinforced learning from human feedback (RLHF).

In short, AI models with human feedback are superior to AI models without it. The more RLHF you have, the better the model. Capturing human feedback can be very expensive. But again, crypto has proven to be good at getting people to do a specific action with token incentives, so we may be able to create systems that capture RLHF in greater scale and in a more cost-effective manner than traditional methods.

Instead of having one company to raise a ton of VC money to fund RLHF, a network can finance this with clever cryptoeconomics.

Crypto may also be crucial to minimize data selection for training bias (data is open, accessible and immutable in blockchains) and platform risk.

Look at Google’s Gemini as an example — one platform has many opinions on the data on which the models are trained. The bet is that companies/developers will want certainty that the models they are utilizing are as unbiased as possible and have cryptographic proofs that models are doing what they are being told.

The most exciting thing about this is that the impact crypto can have on AI is already being felt.

AI will be explosive. Its effect will be felt by everyone, but incumbents may capture even more value. Crypto offers an interesting and viable solution to counteract the centralizing force of technology and restore the original principles of the internet. I see a world where crypto — specifically, this idea of creating participatory networks — will become the de facto go-to-market mechanism for networks.

Crypto is the extension pack of the internet.

— Santiago Roel Santos

Data Center

• Bitcoin’s price is up over 3% to $63,000 early Friday.
• Marathon mined 2,811 BTC last quarter, a decrease of over 30% from the previous quarter.
• Pump.fun has brought in nearly $500,000 in revenue over the past 24 hours, per DefiLlama .
• Grayscale’s outflows have picked back up, with $43 million leaving the fund yesterday, according to data from Farside .
• ETH’s price is up 2% to over $3,000, per Coinbase .

Not selling is still holding

BlackRock ’s memecoin balance is piling up.

Nearly $200,000 in random tokens are parked in addresses linked to BlackRock’s onchain money market fund, BUIDL. New deposits come in every day.

To be clear, BlackRock is not buying memecoins, despite the spoofed Uniswap trades that suggest it had bought trillions of BOMEPEPE.

It’s long been tradition for memecoin holders, and maybe even the creators themselves, to direct funds to the wallets of popular figures.

Read more from our opinion section: BlackRock clearly wants to take crypto seriously. Too bad it’s forever silly.

Think of it as part hat-tip, part marketing spend — the onchain equivalent of renting a billboard inside a blockchain explorer.

BUIDL’s onchain history has seen more than 500 transactions but only about two dozen appear to be actioned by BlackRock itself.

BlackRock initially funded BUIDL with 1 ETH from Kraken and 200 USDC minted directly via Circle. The fund then sent small amounts of USDC between itself, in early-to-mid March, apparently to warm up, about a week before the fund was formally announced.

Since then, Larry Fink’s firm, the world’s largest asset manager, has been interacting with Ondo Finance, which has been acquiring BUIDL’s tokens as one of its reserve assets (there are currently 13 holders of BUIDL).

All while almost 300 different tokens have flooded BUIDL’s addresses, not just on Ethereum, but on any EVM-compatible chain with an analogous wallet, including Binance Smart Chain, Polygon and Avalanche.

Most of the tokens sent to BlackRock to date are worthless. But there are some with active markets, which gives them real value, and not all are strictly memecoins .

Caution is advised when looking into any of these tokens — their place in BlackRock’s wallet isn’t exactly an endorsement.

BlackRock has been forced to hold more than $60,000 worth of UBXS alongside its legitimate stablecoin balance. UBXS is pitched as a utility asset for tokenized real estate app Bixos.

There’s also nearly $30,000 of ISP, which supposedly powers open-world, play-to-earn metaverse platform Ispoverse that can support virtual hackathons.

Although, I’m not sure it really helps perception of those projects to be placed alongside tokens like FINKWIF, NOODS, PEEON and BOBO MUMU.

Still, BlackRock hasn’t sold any of the tokens playing stowaway in its Ethereum wallet. Does that count for anything?

Probably not. Sad.

— David Canellis

The Works

• Kraken is pushing harder for a dismissal in its SEC lawsuit, Bloomberg writes.
• Token projects are increasingly turning to online influencers for help, according to CoinDesk’s Danny Nelson.
• Neel Somani, founder of Ethereum scaling startup Eclipse, says he is “reducing” his role amid sexual misconduct allegations made against him.
• Espresso and Polygon Labs are partnering on rollup interoperability.
• US airlines are expanding their use of AI, according to the New York Times.

The Morning Riff

Pump.fun is exactly what Gary Gensler and the SEC are afraid of.

It’s also one of those things that is so unique to crypto — mixing fun with the potential to profit.

But it’s also risky, as Blockworks reporter Jack Kubinec pointed out in the Lightspeed newsletter yesterday. Folks who go to casinos are probably of the same mindset as a lot of folks on Pump.fun, and for them, the risk is worth the potential reward.

For the SEC and Gensler , though, this is a scenario they claim they want to protect investors from. While Pump.fun says it prevents rug pulls or exit scams, it can’t stop shady people from doing shady stuff.

So while the SEC continues its crusade against companies that have said they want to — or have tried to — come in and register, I’ll sit back and watch crypto’s latest venture play out.

— Katherine Ross

Start your day with top crypto insights from David Canellis and Katherine Ross. Subscribe to the Empire newsletter .