In light of the current market buzz and ample funding, it’s important to acknowledge that we’re in an inflated industry, which may not accurately represent the true state of generative AI. It’s natural to feel perplexed. However, by taking a step back from the hype and examining the Web3-AI landscape based on current needs, it becomes evident where Web3 can bring significant value to the next phase of generative AI. Navigating through this thick veil of misinformation can be challenging.

Web3-AI Reality Distortion

As people deeply involved in cryptocurrencies, we strongly believe in the importance of decentralization in various aspects. However, artificial intelligence (AI) has become more and more centralized when it comes to data processing and computational power. To make a compelling case for decentralized AI, we must first address this natural tendency toward centralization.

In the realm of artificial intelligence (AI), there’s a growing disparity between the perceived value we derive from AI in Web3 and the actual demands of the AI industry. The problematic truth is that the chasm between AI applications in Web2 and Web3 is expanding instead of narrowing, primarily due to three essential causes:

Limited AI Research Talent

A modest number of AI specialists, likely under ten, are currently engaged in the field of Web3. This figure may not instill great confidence in those advocating for Web3 as the next major frontier in artificial intelligence.

Constrained Infrastructure

It’s currently challenging to make web applications function seamlessly with Web3 backends. Therefore, considering AI applications in this context seems premature at best. The computational limitations imposed by Web3 infrastructure prove too restrictive for the development and implementation of generative AI solutions over an extended period.

Limited Models, Data, and Computational Resources

As an analyst, I’ve observed that generative AI depends on three essential elements: models, data, and compute resources. However, none of the cutting-edge models can currently operate on Web3 infrastructures. Additionally, there is a lack of substantial training datasets to build upon in this domain. Moreover, there exists a significant disparity in quality between Web3 GPU clusters and those necessary for pretraining and fine-tuning foundational models.

As a crypto investor and observer of the tech landscape, I’ve come to acknowledge the challenging truth: Web3 has been attempting to develop an “economical alternative” to AI, striving to replicate the capabilities of Web2 AI but falling short. Yet, this reality starkly contrasts with the significant value proposition that decentralization brings to the table in various aspects of AI.

As a researcher exploring the Decentralized Artificial Intelligence (DAI) landscape, I’d like to shift gears from theoretical discussions and delve into specific trends in this field. By evaluating each trend based on its market potential, we can gain a clearer understanding of where DAI is headed.

In the realm of Web3-AI, an illusory effect has influenced the first wave of investment and development towards projects that appear disconnected from the genuine demands of the AI industry. Simultaneously, there are burgeoning sectors within Web3-AI that boast significant promise.

Some Overhyped Web3-AI Trends

Decentralized GPU Infrastructure for Training and Fine-Tuning

Over the past few years, there has been a significant rise in the number of decentralized GPU infrastructures aimed at making it possible for more people to pretrain and fine-tune foundation models. This movement is intended as a response to the dominance of incumbent AI labs in accessing powerful GPUs. However, it’s essential to remember that training large foundation models demands massive GPU clusters with lightning-fast communication links between them. A single pretraining cycle for a 50B-100B model in a decentralized AI setup might take over a year or even fail to execute successfully.

ZK-AI Frameworks

The notion of integrating zero-knowledge (zk) computations and artificial intelligence (AI) has given rise to intriguing concepts for implementing privacy features in foundation models. With the growing significance of zk infrastructure in Web3, several proposals aim to incorporate zk computations into foundation models. However, zk-AI models face a significant challenge in terms of affordability as they can be computationally expensive when applied to large models. Moreover, zk integrations may restrict interpretability – a key aspect in generative AI.

Proof-Of-Inference

In the realm of cryptocurrencies, the focus lies on cryptographic verifications. At times, these verifications are unnecessarily linked to items that don’t require them. In the Web3-AI sector, we come across instances where frameworks generate cryptographic proofs for particular model outputs. However, these situations present more market-oriented difficulties rather than technical ones. Essentially, “proof-of-inference” is a potential solution searching for an applicable use case and remains largely ineffective in today’s market.

Some High Potential Web3-AI Trends

Agents with Wallets

In the realm of generative AI, agentic workflows represent an intriguing development with substantial implications for crypto. Agents here refer to advanced AI systems that go beyond merely providing answers based on inputs; they can also actively engage in executing tasks within a specific environment. Most autonomous agents have been designed for singular use cases. However, the rapid advancement of technology is giving rise to multi-agent settings and collaboration.

In this domain, cryptocurrencies have the potential to generate significant worth. Consider the situation where an intermediary must engage other intermediaries to finish a job or pledge value as a guarantee for the quality of their results. Equipping these intermediaries with financial infrastructure based on crypto facilitates numerous possibilities for collaborative agency.

Crypto Funding for AI

As a researcher in the field of generative artificial intelligence, I’ve noticed that there’s been a significant funding crunch in the open-source AI community. Most labs are no longer able to sustain large-scale projects due to financial constraints. However, the crypto world offers an intriguing solution with its efficient methods of capital formation through mechanisms such as airdrops, incentives, and points. This concept of using crypto funding rails for open-source generative AI is a promising area where these two trends intersect.

Small Foundation Models

Last year, Microsoft introduced the concept of Small Language Models (SLMs) following the launch of its Phi model, which boasted under 2 billion parameters yet surpassed the capabilities of larger language models in computer science and mathematical tasks. The significance of SLMs with a parameter range between 1-5 billion lies in their potential to drive the feasibility of decentralized AI and expand possibilities for on-device AI. Given the current technological limitations, it’s nearly impossible to decentralize multi-hundred-billion-parameter models. However, SLMs are well-suited to run on various Web3 infrastructures, paving the way for substantial value creation with Web3 and artificial intelligence.

Synthetic Data Generation

I, as an analyst, recognize that data scarcity poses a significant challenge for the latest generation of foundation models. In response, there’s growing interest in developing synthetic data generation methods using these models to supplement real-world data. Leveraging the mechanics of crypto networks and token incentives could potentially bring together a vast number of contributors to create new synthetic datasets.

Other Relevant Web3-AI Trends

In the realm of Web3-AI, several intriguing trends are worth exploring beyond proof-of-human outputs. The relevance of human-verified outputs grows in response to challenges posed by AI-generated content. Trust and transparency are crucial elements that Web3 technologies can bring to the evaluation and benchmarking segment of AI. Additionally, human-centric fine-tuning methods like reinforcement learning with human feedback (RLHF) offer an intriguing scenario for Web3 networks. As generative AI continues to advance and mature, new Web3-AI applications are sure to emerge.

The significance of enhancing AI capabilities with a more decentralized approach is undeniable. Although the Web3 industry may not have reached the monetary worth produced by the large AI models, it holds immense potential for the generative AI sector. However, the major hurdle in advancing Web3-AI could be surmounting its own “reality distortion field.” There is great value to be derived from Web3-AI; we merely need to concentrate on creating tangible applications.

As someone with extensive experience in the cryptocurrency industry and a deep understanding of the complexities involved, I want to clarify that the perspectives expressed in this article are my own. They do not represent the views of CoinDesk or any affiliated entities.