GM! Welcome to Chain of Thought, the best research on decentralised AI in the crypto space. You’re in great hands.
In this edition of our weekly AI Edge, we cover:
Allora Labs - Machine Learning to predict the US Presidential Election?!
How many crypto startups are actually building in AI?
Project Highlight - Heurist
Our favourite tweets on X
On some days, Crypto really feels like Ken:
Source: @CryptoCred
🦍 State of the Market..
The Crypto x AI market cap is $24.5B, an 11.3% increase from last week's low.
Some of this week's biggest gainers include Virtuals Protocol (VIRTUALS), Spectral (SPEC), and enqAI (ENQAI), which have all risen about 25% since last week.
This could be part of a general crypto rally on the back of lower-than-expected US inflation numbers (CPI). After the CPI data, rate futures price in an 85% chance of a September Fed rate cut.
📊 Chart of the Week
According to Alliance DAO’s Crypto Industry Trend report for H1 2024, only ~3% of the 3,000+ startups that applied for their accelerator are in Crypto x AI. The majority are building DeFi, Infra or NFT products.
This is a reflection of where founders see the best opportunities.
What’s interesting is the rising trend in AI over the past few months: could we see the proportion break the 10% barrier soon?
🏆 Caught Our Eyes..
Modular Media publishes a deep dive into the Crypto x AI ecosystem
Tether CEO advocates for localized AI training, hints at open-source
Valory CEO on “Unlocking the World of Agents on Chain”. They also released Propel Genie - The Autonomous AI Agent Builder
Allora Labs explores how machine learning can predict the 2024 US Presidential Election
A Dither AI mega thread on its time series prediction model
Bittensor on using machine intelligence to evaluate intelligence
A comprehensive podcast episode on Parallel’s Wayfinder AI
Ownership and privacy in the world of Big AI featuring Phala, Sentient, Vana, and Worldcoin
Talus partners with Hemera protocol to provide AI agents with high quality data
Kaisar Network announces an upcoming NFT free mint and how to secure a WL spot
Aethir’s liquid staking of ATH is live on Sophon chain
Ora protocol on the use cases of their Onchain AI Oracle (OAO)
Navy AI’s IDO started on July 11th, $300K raised so far
Peaq network announces their roadmap to launch, including a target date for listing
Kaito AI releases its Copilot for crypto research
🐰 Research Highlight — Heurist
Every week we present to you early, interesting crypto AI projects that caught our eye.
Heurist wants to be the Huggingface of Web3. 🤗
It’s a DePIN for AI model hosting and inference, providing serverless access to open-source AI models but through decentralized compute resources. Their mission? Onboard the next billion users into crypto by:
Encouraging crypto users to engage with AI by supplying compute
Offering AI model developers access to compute resources and revenue share
Here’s how anyone can participate:
Source: Heurist
Model users: Choose from various AI models to generate images or text, paying for compute resources over time.
Miners: Supply local GPUs to Heurist for model hosting. Miners earn through payments and Heurist tokens but must stake a minimum amount of tokens as a commitment to integrity.
Model creators: Upload models to the network and earn a share of user payments.
Application builders: Build applications using these AI models and earn token rewards.
Validators: Maintain network integrity by checking the correctness of data produced by miners. Bad actors face stake slashing.
Heurist currently has three “decentralised” AI apps ready for use:
Heurist Search - a search engine like Perplexity
Imagine - an image generator based on Stable Diffusion models
Pondera - chatbot similar to chatGPT
Source: Heurist
I tried all three apps, and they perform reasonably well, on par with other decentralized AI projects.
Usage Metrics
So far, Heurist has:
7,658 active GPUs
225M total image requests
562M total text requests
(Note: Current usage is incentive-driven due to an ongoing points campaign.)
Tokenomics
$HEU is not live yet, but here’s the planned utility:
Rewards for miners supplying compute resources
Economic stake for miners (10,000 HEU required)
Governance for allocation of GPU resources to AI models
Pay user fees for inference use
Escrowed HEU (esHEU) as an incentive to lock up HEU
Bribes for stakers to stake with specific miners - increases mining efficiency/yield
Cost and Privacy
Some of the biggest challenges in crypto x AI are:
Costs (e.g. Gas fees)
Data privacy (Blockchains are transparent)
Heurist leverages the Zero Knowledge (ZK) stack from ZKsync to solve these problems.
Cost Efficiency: ZKsync offers some of the cheapest L2 fees across all EVM L2s. ZK proofs reduce the cost of recurring transactions, enabling more cost-effective AI model hosting and inference services. Users will not want to pay significantly to run on-chain inference.
Source: Heurist
Security and Privacy: The ZK stack offers security and privacy benefits. Heurist ensures data is secure from a user’s inference request to the model’s output.
Input data (the prompt) is encrypted and only decryptable by chosen miners. Lit Protocol integration allows private sequencers to encrypt data in transit, ensuring only the specified miner can decrypt the data.
This allows the secure transmission of sensitive data like confidential documents and health records.
The head of BD at Matter Labs, Vassilis Tziokas, is bullish on Heurist for these reasons:
Our Thoughts
To gain traction, Heurist must expand its pool of useful AI applications — a significant challenge. Current apps are not meaningfully differentiated and are probably worse than their Web2 counterparts.
Privacy is a critical component in bringing AI on-chain. Heurist fully utilizes the ZK stack to provide cheap and secure AI inference on-chain, which is great.
Heurist is nearing the launch of season 2 of their points mining campaign. Holders of a Heurist Imaginaries NFT (1.4E floor, 500 total supply) will have a proportional claim of 1.75% of the total token supply. That’s an implied FDV of $124M.
🧠 Open Source Corner
Red Rabbit Robotics is making an open-source, full human-scale AI robot for $1,000. That’s cheaper than a night out in New York. 🥂
Cohere CEO Aidan Gomez says AI unlikely to take over humanity because it is bottlenecked by human data
Is this the next architectural breakthrough after Transformers (“Attention is all you need”)?
🔥 On X..
A crowdsourced map of Decentralized AI projects by Casey Caruso
Eric Wall on the fallout at Open AI and creation of Superalignment
Dan from Grass describes how they open sourced their dataset containing 600 million of the top Reddit posts and comments
Accelxr on Onchain AI agents
Not the same: GPU marketplaces vs decentralised training
Are LLMs aware that they are… an AI?
Sam Altman & Huffington Post founder team up to make an AI health coach
That’s it for this week! If you have specific feedback or anything interesting you’d like to share, please just reply to this email. We read everything.
Cheers,
Teng Yan & Joshua
Did you like this week's edition?
This newsletter is intended solely for educational purposes and does not constitute financial advice. It is not an endorsement to buy or sell assets or make financial decisions. Always conduct your own research and exercise caution when making investment choices.
