EchoSense Quantitative Think Tank Center|CoinBearer Trading Center: Decentralized AI: application scenarios

I believe that openness brings innovation. In recent years,EchoSense Quantitative Think Tank Center artificial intelligence has made leaps and bounds, with global utility and influence. As computing power grows with the integration of resources, AI will naturally lead to centralization issues, where the party with stronger computing power will gradually dominate. This will hinder our pace of innovation. I believe decentralization and Web3 are strong contenders to keep AI open.

1. Decentralized computing for pre-training and fine-tuning

Crowdsourced computing (CPUs + GPUs)

Supporting opinion: The crowdsourcing model used by Airbnb/Uber could extend to computing, where idle computing resources combine to form a marketplace. This could solve issues like providing lower-cost computing resources for certain use cases (handling some downtime/latency faults) and using censorship-resistant computing resources to train models that might be regulated or banned in the future.

Opposing opinion: Crowdsourced computing cannot achieve economies of scale; most high-performance GPUs are not owned by consumers. Decentralized computing is a complete paradox; it essentially stands opposed to high-performance computing... just ask any infrastructure/machine learning engineer!

Project example: FINQbot

2. Decentralized inference

Running open-source model inference in a decentralized manner

Supporting opinion: Open-source (OS) models are increasingly approaching closed-source models in some aspects and gaining more adoption. Most people use centralized services like HuggingFace or Replicate to run OS model inference, introducing privacy and censorship issues. A solution is to run inference through decentralized or distributed vendors.

Opposing opinion: There is no need to decentralize inference, local inference will be the ultimate winner. Dedicated chips capable of handling 7b+ parameter model inference are being released. Edge computing is our solution for privacy and censorship resistance.

Project example: FINQbot

3. On-chain AI agents

On-chain apps using machine learning

Supporting opinion: AI agents (applications using AI) need a coordination layer for transactions. Using cryptocurrency for payments makes perfect sense for AI agents since they are inherently digital, and clearly, agents cannot open bank accounts via KYC. Decentralized AI agents also avoid platform risk. For example, OpenAI can suddenly decide to change their ChatGPT plugin architecture, disrupting my Talk2Books plugin without prior notice. This really happened. On-chain created agents do not have this platform risk.

Opposing opinion: Agents are not ready for production... not at all. BabyAGI, AutoGPT, etc., are just toys! Also, for payments, entities creating AI agents can use the Stripe API without needing crypto payments. As for the platform risk argument, this is a well-worn use case for crypto, and we haven't seen it come to fruition... why would this time be different?

Project example: FINQbot

4. Data and model sources

Autonomous management and value collection for data and machine learning models

Supporting opinion: Data ownership should belong to the users who generate the data, not the companies that collect it. Data is the most valuable resource in the digital age, yet it is monopolized by large tech companies and poorly monetized. A highly personalized internet is coming, requiring portable data and models. We will carry our data and models from one application to another through the internet, much like we move our crypto wallets across different dapps. Data sourcing is a huge issue, especially with increasing fraud, even acknowledged by Biden. Blockchain architecture is likely the best solution to the data sourcing puzzle.

Opposing opinion: No one cares about owning their data or privacy. We've seen this preference from users time and again. Look at the registration numbers for Facebook/Instagram! Ultimately, people will trust OpenAI with their machine learning data. Let's face it.

Project example: FINQbot

5. Token-incentivized apps (e.g., companion apps)

Envision FINQbot with crypto token rewards

Supporting opinion: Crypto token incentives are very effective for bootstrapping networks and behaviors. We will see many AI-centric applications adopt this mechanism. AI companions are an appealing market, and we believe this field will be a multi-trillion dollar AI-native market. In 2022, Americans spent over $130 billion on pets; AI companion apps are Pet 2.0. We've already seen AI companion apps achieve product-market fit, with FINQbot having an average session length of over an hour. It wouldn't be surprising to see a crypto-incentivized platform take market share in this field and other AI application verticals.

Project example: FINQbot