Road to Sustainable AND Profitable Open-Source AI

Most models are Closed-Source. Shifting towards an Open-Sourced ecosystem requires overcoming Data Quality, Business Model, and Funding problems.

THE PROBLEM:

Open-source AI models don't get the same amounts of compute and capital as closed-source AI models are getting.

Current Open-Source Business Models are Unsustainable

Most models are closed-source.

The ways models are monetized today are:

1. Serving more prominent companies’ existing business (Meta, OpenAI, Google)

2. SaaS subscription models (OpenAI, Anthropic)

Open-Source Models :

• Open Hermes 2 ( by nous research)

• Zephyr 7b

• open chat

• Mistral

• Llama

• Yi (by 01.AI)

• Qwen

---

The Funding Problem with Open-Source AI:

Open-source AI needs to find ways to monetize so developers can work on fine-tuning models and releasing them to the world.

Some Approaches for Making Open Models Sustainable:

1. Building evaluation methods for datasets - Datasets are today's most crucial resource in new models. As more models come online, this presents an opportunity for millions of developers worldwide to create specialized models and build businesses around them. Developers today find it challenging because it is hard to demonstrate the quality of the datasets they make. Building evaluation standards and benchmarks for open models is critical to a Cambrian explosion of open-source models.

   • Potential business models include:

     1. Keeping the datasets closed-source and licensing them while the model remains open.

     2. Tokenizing the model.

2. Tokenizing the ownership of the model to the people who use it - We believe this is the holy grail of open-source AI. Tokenize models.

Tokenized Models:

What we envision is a model that developers use to build apps. They pay the model for API calls and, while paying, earn an ownership stake in it.

What Is the Business Model of a Model? We imagine a business model similar to a Layer 1 (L1) blockchain, where the model is used to build decentralized applications (DApps) and charges 'gas'. Applications created this way generate revenue, some of which return to the model.

What the Token Will Do:

The token will represent ownership and be distributed to users based on usage. The more you pay, the more ownership you acquire. This ownership entitles you to a share of the revenues from the built apps.

The Potential of the Combination of Evaluation for Datasets and Tokenization for Models:

With this combination, it will be MUCH easier for anyone fine-tuning an open-source model to secure funding because:

1. The model will generate money predictably and straightforwardly - users are owners, and the compute cost will always be covered (by the users).

2. The assurance of obtaining high-quality datasets will be more significant - so investors will be confident that the model will be of high quality. In addition, the dataset alone will be an asset that the developer can potentially monetize, possibly by licensing it to others.

The World We Imagine:

We envision a world with millions of specialized models and millions of young, millionaire open-source developers who have created something hugely valuable while working from their parent's homes. We want to encourage this world.