Generative AI: revolutionizing blockchain?

Generative AI, a rapidly evolving technology, has sparked seismic changes across the tech spectrum, with its influence pervading everything from infrastructure to applications. In recent years, the potential integration of this transformative tech with the Web3 ecosystem has sparked fervent debate, exploring innovative applications such as AI-driven digital wallets and language exploration tools.

A compelling question arises: does generative AI necessitate its proprietary blockchain?

Balancing the scale: open-source progress against centralized governance

The proposition of a dedicated blockchain for generative AI warrants a review of current advancements and concerns surrounding foundation models, especially the growing open-source alternatives to API-based technology like GPT-4 and the ongoing debates over centralized control.

In the recent past, API-based models like OpenAI’s GPT-4, Anthropic’s Claude, DALL-E, and Midjourney held a distinct advantage over their open-source counterparts in terms of their quality. However, the tables began turning with the unexpected open-source release of Stable Diffusion, an alternative to API-based text-to-image models, triggering a significant shift towards open-source models.

Adding fuel to this open-source fire, Meta AI Research presented a paper on LLaMA, a large language model (LLM) with the capability to compete with GPT-3, despite its significantly smaller size. While the initial plan was to keep LLaMA proprietary, an accidental leak resulted in a mass download and an unprecedented acceleration of open-source innovation.

This momentum gave rise to numerous open-source foundation models such as Alpaca, Dolly, Koala, and Vicuna, significantly tipping the scales in favor of open-source generative AI. As the quality gap between open-source and commercial models continues to close, concerns around centralized control and transparency within the AI industry intensify.

These twin forces — the growth of open-source generative AI and the mounting concerns over centralized control — offer a unique opportunity for the integration of blockchain in AI.

Charting a new course: generative AI on Web3

The open-source revolution has made it significantly easier for Web3 platforms to adopt generative AI capabilities. Two potential paths present themselves:

• The development of decentralized applications (DApps) leveraging generative AI tools.
• The creation of new Web3 platforms with generative AI at their core.

In the first instance, existing tools could incorporate AI-driven features, leading to a new generation of DApps. However, this would primarily have Web3 as a user of generative AI, running the models on traditional cloud infrastructures.

The real game-changer would be Web3 platforms that can inherently support generative AI models. A blockchain explicitly built for generative AI could then be within our grasp.

Building a new blockchain optimized for generative AI might be challenging but it isn’t without precedent. Given the architectural complexities of running foundation models on current blockchain runtimes, it stands to reason that we might need to design a unique infrastructure to accommodate them. The core functionalities of this new blockchain could include executing foundation models, running pretraining and fine-tuning workflows, and sharing datasets used for training models.

Addressing the need for a new blockchain

The prospect of a dedicated generative AI blockchain is enticing, but is it essential? Generative AI has the potential to reshape the blockchain infrastructure. It’s feasible to imagine a proof-of-stake blockchain with transactions processed in natural language or smart contracts utilizing language as the primary means of communication.

Ignoring the integration of generative AI in Web3 could result in a widening technological divide between current and future web architectures. Generative AI is altering the fundamentals of software development, and new platforms supporting this paradigm shift are emerging rapidly. Therefore, integrating native generative AI capabilities into Web3 isn’t merely desirable; it’s a crucial step towards ensuring its future relevance.

While creating a new blockchain presents numerous challenges, the fast-paced advancements in Layer 2 runtimes, platforms like Cosmos, and the rise of high-performance Layer 1 ecosystems like Aptos or Sui make the prospect of a generative AI blockchain more attainable than ever.