For the last couple of months, we keep getting exposed to interesting ideas and startups in the intersection of AI and web3, and some of the ideas make a lot of sense to us.
First idea, a decentralized data marketplace. Companies are increasingly requesting AI models trained only on licensed data – a driving force of partnership between model companies and content publishers. Reddit made the news for earning $60M on user data, but none of the users saw this. You can imagine a protocol on top of something like Farcaster (web3 social platform) where users opt-in to sell their data, and when AI companies license this, the money is split between users. Slightly different but relevant – Numbers Protocol is building content verification via a provenance infrastructure to label and track human vs AI generated images and more.
Decentralized compute has been tried over the years – this feels promising but there are challenges such as latency, etc. Bagel is a news startup taking a stab at decentralized hosting of data and model compute. Another interesting approach is decentralized internet access, like Grass – who uses a network of decentralized computers with internet access to scrape sites. They recently open sourced 600 million Reddit posts and comments collected this way.
This leads to decentralized knowledge gathering – i.e. a web3 version of Wikipedia. Combining this with our love for graphs, you get something like Index Network or OriginTrail for building and sharing knowledge graphs. Sapien has a data labeling game incentivized by blockchain rewards. You also have groups privately collaborating, such as the quant community of data scientists collaborating on ML models at CrunchDAO.
This one is less obvious, but Gen AI media IP might work well with web3. One observation from watching our kids play with Gen AI tools is that they keep asking for characters they know and love – which many tools won’t produce due to licensing. As one example, a variation of NFT tech could be used to handle permissions within AI tools on what characters you can generate and with what items.
Finally, you have web3 for AI agents – think identity, payment, and collaboration. You know about PayMan which I mentioned above, which can do both fiat and crypto payments. ChainAgent and Altered State Machine are for building on-chain agents. Then you have Sahara Labs who raised a large $43M round from Sequoia to build a suite of AI-native blockchain tools including a personalized agent.
Leave a Reply