## AI on NEAR: A Look Ahead As many know by now, my background before founding NEAR was in AI research (TensorFlow, Google AI, coauthor of “Attention Is All You Need,” the paper that introduced the model architecture powering GPT-4 and other major LLMs). The recent surge of interest around Chat-GPT and other LLMs has sparked interesting conversations with the media and ideas for the intersection of AI and blockchains. I see several ways this intersection can start to happen here on NEAR and want to share some plans for how we’ll do this in the Blockchain Operating System. We will build some of this at Pagoda in the second half of the year and I hope to see ecosystem projects start to form around these, as well. At a high level, blockchains can help coordinate decentralized groups of contributors around doing some work according to a set of rules, so they have a role to play in governance of AI as well as globally sourcing contributions to training models. TLDR: open source always wins. We see this throughout the history of tech products, and I don’t think AI will be any different in the medium to long term. Blockchains can help make AI more open while making it easier to proactively monitor bias, authenticate data, and manage safety protocols. The potential power of this technology is amazing, but so are the potential risks, and the best way to mitigate that risk is to build more open systems for managing them. We can integrate the GPT-4 API into the BOS and bring those interactions on-chain, introducing cryptographic signatures to every interaction with the model. At a high level, these are the AI+blockchain solutions that are already possible to build today inside the BOS. * __Crowdsourced data labeling__: a marketplace to reward contributors for high quality data labeling that trains a model - Globally sourcing this work scales up training capacity while also reducing potential bias - Already happening on NEARCrowd, will keep expanding on this approach * __Governance__: managing and monitoring AI transparently and proactively - Creating alignment dataset via crowdsourcing - Curating public datasets for evaluation of models - Directing research funding - Public and traceable research discussions (vs twitter flame wars) * __Content management system & on-chain reputation__ - Introduce provenance tracking of all media or content assets when they are created: text, articles, a quote, images, videos - Solves problem of fake attributions, misaligned or illicit content - Authenticated content provides traceability as users interact with data over time - Traceability also offers a more powerful user control over content they want to see with community governance and on-chain reputation system * __Language defined experiences__: a user can ask an AI to create a frontend for them combining different components and features. Automating this process will be extremely powerful for making development accessible to pretty much anyone with an account. This is nearly impossible to do in Web2 because all the source code and services are closed. Fun fact: Alex Skidanov and I were exploring exactly this approach to generative app creation in the year before founding NEAR––it wasn’t possible then, but now it is. generating frontends in BOS via user description connecting with APIs and smart contracts * __Marketplaces__: can make interaction with LLMs and deep learning models overall more accessible while still allowing monetization around models that are expensive to run and train - Hardware: access to GPU clusters powering LLMs, improve price discovery and lower the barrier to entry for researchers - Datasets: introduce a market for high quality data contribution or requests - Trained models: access to LMMs and other models * __Summarization component__ - Maintain summary of posts / topics in social - Quickly glance over the governance discussions in DAOs - Summary of the content under link * __Text suggestion component__: - Suggesting next content to post, help edit proposals in the DAOs There are a lot more use cases between BOS and AI. To keep up with this progress, keep following along on NEAR Social and Twitter for more updates and before too long, some launches.