From OpenAi to Bittensor: A Paradigm Shift for Decentralized AI Networks

Written by Teng Yan, Head of NFT Research, Delphi Digital, translated by xiaozou

One of the things you have to get used to living in Asia is that you often wake up to big news and have to do your homework to stay behind.

For example, Sam Altman was fired from OpenAI last Friday, and I almost choked on milk when I saw the news.

Why would the board of directors fire an extremely smart success story who just gave a brilliant keynote at OpenAI 12 days ago?

Andrew Cote believes that Altman was fired for political reasons because “he may be pushing AI too fast by deploying the latest breakthroughs.” "Some people don’t like that.

OpenAI’s company structure is very strange (almost dysfunctional) because OpenAI started as a non-profit organization and later decided to transform into a for-profit business. Today, nonprofits control the direction of for-profit entities while providing limited upside for investors.

The next few weeks will be very exciting as the truth comes out.

Will this be another Steve Jobs moment, and will Sam start another company to compete with OpenAI?

But what is certain is that OpenAI’s inner workings are shrouded in mystery. Although GPT has become a ubiquitous tool and is used by hundreds of millions of people around the world, there is still a clear disconnect.

As regular everyday users, we find ourselves standing outside, trying to peek in through the veil of mystery surrounding these AI giants. As GPT continues to be integrated into every aspect of our society, this lack of transparency is worrying.

Lately, I’ve been thinking about what the intersection between crypto and AI looks like, but most people agree that such a convergence would unlock great potential.

When we think of AI x Crypto (the convergence of AI and crypto), we usually think of Akash Network and Render. These are GPU decentralized networks that can provide the necessary computation for AI model training. The logic is simple – as AI continues to skyrocket, so does the demand for computing resources. In this case, there could be significant growth in peer-to-peer networks. So they’re in the picks and shovels business, but I think that’s only scratching the surface of the potential of AI x Crypto.

It’s like saying that monkey JPEG is the pinnacle of NFTs.

Then I came across Bittensor.

#1 Hell5:Abundancesor

Unlike Akash or Render, which support AI model training (upstream), Bittensor focuses on AI inference (downstream), using the trained model to generate output.

Bittensor is a decentralized network that incentivizes AI models, specifically large language model LLMs, to handle a variety of tasks such as text generation, image creation, and music production. Currently, the network has 27 subnets, each focused on a specific task.

To put it simply, think of Bittensor as anything that a decentralized ChatGPT + Midjourney + AI can do. **

The network operates through two main roles:

Miners (Value Producers): Miners develop and host AI models on the network. Based on the performance of the model in relation to a specific task, they will be rewarded with TAO tokens. This incentivizes the development of better and more efficient AI models. Validators (Consensus Producers): Validators evaluate the output of miners, ranking their performance on specific tasks. They also interact with users who submit tasks to validators and send them to the appropriate miners.

I may have oversimplified the technical complexity, but a few things are obvious to me:

• Miners and validators on the network exchange knowledge and share parameters, which can self-optimize over time.
• The network is designed to leverage the strengths of multiple independent AI models to produce the best possible output (“expert set”).

#2 T

TAO is the utility token of the Bittensor network, which is similar to Bitcoin’s tokenomics structure: a hard cap of 21 million tokens and a fair release, with no VC allocation. It even has a halving cycle, with the first halving taking place in 2025.

Today, there are 5.65 million TAO in circulation, all of which are fairly distributed through mining and verification on the network. TAO’s current outstanding market capitalization is just over $1 billion. The number of new TAO released to miners and validators every day is 7,200.

3, a little bit of my thoughts

Bittensor is still in its initial stages. The network has a devout community, but the size of the participants is still small – just over 50,000 active accounts. The busiest subnet, SN1, is dedicated to text generation, with around 40 active validators and more than 990 miners.

What really appeals is the concept of decentralized AI networks, which reduces the risk of centralization while also raising the question: Can these unique economic incentives foster AI models that go beyond those developed by well-capitalized entities like OpenAI and Google?

Before LLMs became mainstream with the advent of tools like ChatGPT, deep tech startups typically focused on acquiring proprietary data to develop specialized, machine learning-based AI models for specific tasks. For example, Flatiron Health uses real-world clinical data from cancer patients to develop AI models, incorporating them into tools to support cancer researchers and health care providers. Historically, startups have aimed to productize and monetize these proprietary models.

However, Bittensor may represent this paradigm shift. Perhaps it would be more appropriate to say that this is a technology-driven business model innovation rather than a technological breakthrough. For example, it provides a path for proprietary data and AI models to be co-developed for a wider audience without the need to open source them. I can envision a future where Bittensor has thousands of specialized subnets that can address a range of challenges, whether it’s environmental, healthcare, or energy issues.

Truth be told, I find it fascinating if a team can design their tokenomics in the same way as Bitcoin. This shows their motivations, which are different from today’s teams, who often optimize their tokenomics in accordance with the venture capital model, offering a large distribution of tokens to founders and investors.

I’m not sure where Bittensor is going. It can be a hundredfold success, or it can fail completely. But its potential and the philosophy behind it are too fascinating for me to remain indifferent.