A Beginner’s Guide to Bittensor

Here's your concise guide on Bittensor network!

Introduction:

The Bittensor Network, created and overseen by the Opentensor Foundation, is an innovative blockchain initiative that seeks to improve the quality of Artificial Intelligence (AI) models via decentralized, incentivized competition.  If you’re a complete beginner to machine learning or blockchain but would like to learn more about this fascinating project leveraging both of these concepts, read on!

Within the Bittensor Network, there are two kinds of nodes: The Miner (which is an AI model), and The Validator (which serves as the rewarding mechanism, as well as the gateway by which users / applications submit queries to the network). The rewards for work are provided by means of a share of the network’s utility token, the TAO, which is mined via Bitcoin-like tokenomics (as a reward for work done on the network, with a similar halving schedule). But instead of racing to be the first to find the answer to a difficult computation (Proof-of-Work), the competition is to provide the highest quality model outputs compared to their peers.  With a recent upgrade to the network, Bittensor has enabled subnets, each of which can be even further designed for a specialty – image generation, text generation, incentivized storage, pre-training, etc.   For the purpose of this article, we will be describing text-generation subnets that were part of the original implementation of Bittensor (and which are currently maintained by the Foundation within subnets of their own).  Please note that the actual function and requirements of subnets devoted to specializations other than text generation will be different from what the article describes, below.

There are two kinds of nodes within the Bittensor Network

Validators operate on a Proof-of-Stake basis: they must stake TAO tokens to become a validator, and their own share of rewards is based on that stake (both the initial stake by the validator’s operator, and the tokens delegated to it by TAO token holders).  Within each validator, there is a “reward module”, which is actually a stack of several AI models in and of itself, which is constantly being tweaked by the Opentensor Foundation to ensure that it is able to properly understand, rank, and reward the answers provided to it by the miners, as well as several other means to ensure that the answer it is receiving is coming from an actual AI model (as opposed to a sybil — a dumb program that listens for specific queries and responds with what its operator has told it is “the best answer”).

Unlike many blockchain projects, the Opentensor Foundation never held an Initial Coin Offering or offered TAO tokens for sale as a fundraiser (or as rewards for community participation).  Nor does the Foundation market TAO, make use of influencers, or seek listings with exchanges.  Instead, the network was simply opened to those who were able to set up miners and validators. Naturally, since the Foundation initiated the chain, its validator holds a commanding amount of stake, as do the validators of other third-parties who were in on the project from the beginning.  But the consequence of that is that the token itself has been relatively shielded from rampant speculation and pump-and-dump schemes; as of this writing, the token seems to hold a solid value and while of course is value does fluctuate with the market, the swings are not as wild and destructive as those tokens whose value is purely driven by speculation. In short, the TAO token has value because of its actual utility.

The recent upgrade to the system allowed third-party developers to set up subnets — the specific partitions that allow the models within them to focus on a particular kind of work – allow model operators to fine-tune their models with extra precision and, because they serve a specific purpose, can be smaller and more efficient. This allows for more rapid training & tuning of models, and ultimately a better product.

What Can End-Users Do with Bittensor?

From the perspective of someone seeking to use AI-driven services, it’s not the Bittensor network itself that is the focus, but rather what the third-party developers are building that make use of the network’s combined model power.  For example:

 – Image Studio (https://studio.bitapai.io/generate): AI-generated images

 – Chat with Hal (https://chatwithhal.vercel.app/): AI chat assistant

 – Reply Tensor (twitter.com/replytensor): AI-generated responses to Twitter/X posts (tag the account in a reply to a post; it will reply to that post with an AI-generated response)

More services for end-users are in active development as a result of the subnet upgrade to the Bittensor network.

Involvement in Bittensor Network Itself

If you are interested in becoming involved in the underlying network itself, there are several ways to do it, ranging from “level 1” to “level 3”.

Level 1: stake to a validator. Once you’ve acquired some TAO tokens, you can delegate (stake) them to a Bittensor validator. Delegates receive 82% of their proportional share of the validator’s overall rewards (18% of an individual stake’s proportional reward goes to the validator as its fee).

Since validators serve as the gateway through which an application accesses the network, validators tend to be run by developers and other groups that are building on the network in some way.  The rewards that are granted to validators are a function of its total stake; therefore, you can support a validator’s rewards (and thereby the rewards to its operators/developers) by delegating your tokens to them.

Level 2: run a validator. This is a more challenging prospect since you will need to keep on top of network updates; by and large, updates happen automatically, but for extremely major changes, the Foundation may require validators to update their code manually.  It also requires a significant outlay of tokens in order to remain competitive on most subnets, so it is not for the faint of heart.

Level 3: run a miner/model. Because Bittensor is a Proof-of-Intelligence blockchain, miners are AI models who are competing against one another to provide high-quality answers to queries. 

This means that in order to mine successfully, you cannot simply load up a pre-trained AI model and walk away from it.  You will need to fine-tune your model, which means you’ll need to have or get some knowledge and experience in the world of machine learning (since you would be competing against people with strong backgrounds in this field).  And you will need to monitor and tweak your model(s) on a regular basis in order to remain competitive.  Lastly, registering a model to the network requires TAO, which you will “recycle” (burn); if your model fails to remain competitive and is deregistered from the network by the consensus process, you’ll have to recycle more TAO to register.

If you are interested in running a validator and a model, however, there is a testnet where you can experiment and learn without needing any mainnet TAO tokens.

Bittensor Network at a glance

Bittensor Resources

Social platforms related to Bittensor are divided into two areas of focus:

The Opentensor Foundation – devoted to development of the network itself and technical education, but does not host conversation about price, market, or investment-related topics

The Bittensor Community – oriented towards the development of third-party applications and does host investment/market-related discussions

Conclusion

In short, Bittensor aims to use blockchain technology to drive innovation in machine learning, via a decentralized, incentivized, competitive process. If you are interested in Artificial Intelligence and its potential uses, and if you are also interested in blockchain technology, then Bittensor may be a project worth investigating!

Recent Posts