Author: 0xJeff, Crypto KOL

Compiled by: Felix, PANews

Today, everyone is selling something, whether it's food, housing, encyclopedias, electronics, applications, or the latest AI.

In the past, what was sold were practical items that satisfied the lower levels of Maslow's hierarchy of needs; today, what is sold are dreams and hopes, packaged in shiny exteriors, especially in the field of crypto AI.

Crypto AI products and infrastructure are often difficult to understand, leading teams to use excessive jargon in communication, failing to attract users.

Moreover, launching a true AI lab (not just simple packaging) requires substantial funding to support talent, contributors, computing resources, and other necessary resources.

Advanced enterprise-level AI labs can cost millions of dollars annually. If cutting-edge AI models are being researched, trained, and optimized, costs can reach hundreds of millions of dollars. The price of H100 GPUs ranges from $25,000 to $40,000, while newer Blackwell B200 and GB200 GPUs range from $30,000 to $70,000. Training a cutting-edge model may require thousands of such GPUs.

Advantages of Decentralized AI (DeAI): Small Models + Reinforcement Learning

Choosing a decentralized system, which coordinates computing resources globally to train a single model, can theoretically significantly reduce GPU costs (saving 30% to 90%), as you can leverage a global network of idle GPUs. However, in practice, coordinating these GPUs and ensuring they all work effectively is very challenging. Therefore, there is currently no decentralized AI lab capable of overcoming the challenges of decentralized training.

However, there is hope for the future, as a few labs have achieved encouraging results in decentralized reinforcement learning. It is this process of self-play and self-learning that can make a small model extremely intelligent.

Not all situations require large language models (LLMs). Training domain-specific models and using reinforcement learning (RL) to refine and enhance their skills is the most cost-effective way to provide enterprise-level AI solutions, as ultimately, what customers want are results (compliance, safety, cost-effectiveness, and increased productivity).

As early as 2019, OpenAI Five defeated the then-world champion OG team in Dota 2. This was not a fluke but a complete domination, winning against OG in two consecutive matches.

You might be curious how it achieved this?

Dota 2 is an extremely complex multiplayer online battle arena game where five players compete against each other to complete various objectives and destroy the opposing base.

To enable the AI to compete with top players, it followed these steps:

• Self-play from scratch: Learning the basics through millions of self-play matches. If it wins, it indicates good moves; if it loses, it indicates poor moves (i.e., large-scale trial and error).
• Setting up a reward system (points) to incentivize behaviors that may lead to victory with positive expected value (EV) (such as destroying towers, killing heroes) while penalizing behaviors with negative expected value.
• Training using a reinforcement learning algorithm called "PPO," where the AI tries certain actions in matches, and PPO treats the results as feedback. If the outcome is good, it does more of that; if the outcome is poor, it does less. This gradually steers the AI in the right direction.
• Hundreds of GPUs running for nearly a year to train the AI, allowing it to continuously learn and adapt to game version updates and changes.
• After some time, it began to explore complex strategies on its own (sacrificing a lane, adopting conservative or aggressive play at the right moments, timing large-scale attacks, etc.) and started competing against and defeating human players.

Although OpenAI Five has been retired, it has inspired the notion that small models can also be extremely effective in domain-specific tasks (OpenAI Five's parameter count was only 58MB).

Large AI labs like OpenAI can achieve this because they have the funding and resources to train reinforcement learning models. If a company wants its own OpenAI Five for fraud detection, factory robots, autonomous vehicles, or financial market trading, it requires substantial funding to do so.

Decentralized reinforcement learning addresses this issue, which is why decentralized AI labs like Nous Research, Pluralis, gensyn, Prime Intellect, and Gradient are building global GPU networks to collaboratively train reinforcement learning models, providing infrastructure for enterprise-level domain-specific AI.

Some labs are researching further cost reduction methods, such as using RTX 5090/4090 instead of H100 to train reinforcement learning models. Others focus on using reinforcement learning to enhance the intelligence of large foundational models.

Regardless of the research focus, it will become one of the most promising development directions for decentralized AI. If decentralized reinforcement learning solutions can be commercially scaled, enterprise clients will invest heavily in AI and will see more decentralized AI teams achieving eight to nine-figure annual revenues.

Funding and Scaling DeAI through Coordination Layers

However, before achieving annual revenues in the eight to nine-figure range, they need to continuously research, implement, and transition to commercially viable reinforcement learning solutions, which requires substantial funding.

Raising funds through coordination layers like Bittensor is one of the best approaches. Millions of dollars in TAO incentives are distributed daily to subnets (startups and AI labs), while contributors (AI talent) contribute to the subnets they are interested in to earn a share of the incentives.

Bittensor enables contributors to participate in AI development and allows investors to invest in AI labs contributing to DeAI technology.

Currently, several key DeAI subfields are emerging in the Bittensor ecosystem, including quantum computing, decentralized training, AI agents, and prediction systems (reinforcement learning is not yet one of them, but there are more than three subnets actively focusing on decentralized reinforcement learning).

What is the current progress in decentralized reinforcement learning?

Reinforcement learning has proven to be applicable on a large scale but has not yet been industrialized. The good news is that the demand for AI agents that can learn from real feedback is rapidly growing. For example, agents that can learn from real-world environments, sales, and customer service calls, and trading models that can adapt to market changes. These self-learning systems can create or save millions of dollars for businesses.

Privacy technologies are also on the rise. Technologies like Trusted Execution Environments (TEE), encrypted embeddings within TEEs, and differential privacy applied in feedback loops help encrypt and protect private information, allowing sensitive industries like healthcare, finance, and law to maintain compliance while having powerful domain-specific self-learning AI agents.

What comes next?

Reinforcement learning is the best choice for continuing to make AI smarter. Reinforcement learning transforms AI from generative systems into proactive, intelligent AI agents.

The combination of privacy and reinforcement learning will drive enterprises to truly adopt and provide compliant solutions for customers.

Reinforcement learning makes the "agent economy" possible, where agents purchase computing resources, negotiate with each other, and provide services.

Due to cost-effectiveness, decentralized reinforcement learning will become the default method for scaling reinforcement learning training.

Federated Reinforcement Learning (Federated RL) will emerge, allowing multiple parties to learn collaboratively without sharing local sensitive data, combining privacy protection with self-learning, greatly enhancing intelligence levels while meeting compliance requirements.

Related reading: Crypto AI Reshuffle: Virtuals Fall from Grace, DeFAI and Predictive AI Seize the Opportunity

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。