The past five years in technology have been dominated by two buzzwords: blockchain and artificial intelligence. While both technologies certainly have value and place in our society, both have likely been over-hyped and underdelivered thus far. That being said, integrating the two technologies — blockchain for trust and transparency and AI for intelligence and reasoning — could prove that the sum is more valuable than its parts. Bittensor, a new project tackling this intersection of buzzwords, is using Bitcoin as inspiration to create a peer-to-peer intelligence market. Let’s dive into Bittensor’s history, what makes it unique, and the value proposition of the TAO cryptocurrency.
Large technology conglomerates like OpenAI, Google, Meta, and Microsoft dominate the current artificial intelligence and large language model ecosystem. They have access to near-infinite amounts of user data and online information, making models like GPT-4 and Google Bard significantly stronger and more well-trained than open-source models.
The Opentensor Foundation created Bittensor to be a blockchain-based machine learning network. It allows different large language models to train together. Miners, or trainers, are rewarded in TAO, the native token of the blockchain, based on how much value they provide. Validators are nodes that measure and value miners’ contributions, helping to decide their TAO distribution. This measurement is done using other intelligence systems, which learn how to value their peers’ contributions and models over time. The idea is to create a decentralized, incentivized language model that, thanks to distributed computing, becomes the winner in a winner-take-all market.
Like Bitcoin, TAO underwent a fair launch with no tokens pre-mined and has a maximum supply of 21 million coins, with 5 million currently circulating. It also incorporates halving events to reduce coin inflation over time. A project tackling this issue in the AI and blockchain space could have raised a lot of money at a high valuation. The fact that they did not is an incredibly positive sign for several reasons. First, it allows users to get tokens at a much lower valuation than is typically possible for a promising project. Second, it signals that the Opentensor Foundation is more interested in creating a good project than earning short-term profits or creating an easy exit opportunity.
In some sense, Bittensor is the natural evolution of Bitcoin, as it builds upon the concept of a digital commodity to create digital markets for artificial intelligence computing power. The energy and resources dedicated to Bitcoin’s mining are purely used to create BTC, while Bittensor’s resources are dedicated to creating and training LLM markets to earn TAO. Then, like people use BTC to pay transaction fees, clients can use TAO to access the Bittensor-trained models.
The Bittensor blockchain is actually a Parachain on Polkadot, meaning it is interoperable with the vast swath of Polkadot chains and can leverage cheap transactions and fast confirmations made possible by Polkadot’s consensus mechanism. This is separate from the digital compute market operation, meaning the computationally expensive data validation can be off-chain.
The LLM space is a winner-take-all environment, meaning the best solution will see most of the market share. Given the massive head start that other LLMs already have, Bittensor is certainly a long-shot bet on distributed computing. That being said, if they can slowly train their models and gain traction, the potential network effects from millions of miners actively contributing to these models could overpower the computing power of even the largest tech companies.
By Lincoln Murr