Author:bebis
Compiled by: TechFlow
Artificial intelligence is really good at disguise.
The development of artificial intelligence (AI) technology has a history of more than 50 years and has attracted hundreds of billions of dollars in R&D investment. Nowadays, it has become exceptionally easy to create an apparently revolutionary AI application, but often these applications may just be "castles in the air".
So how can we distinguish real technological breakthroughs from false hype?
See through the essence of AI marketing
First, we need to understand the reality of software development. Whether it's a base model with trillions of parameters or a small project completed by a developer on a weekend, the process from prototype to production environment is often full of challenges, like going through a "protracted war".
In this "war", the vast majority of the battles often take place in the seemingly final stage. When everything goes smoothly in local testing, when you start to migrate the project to the production environment, you will realize countless times that coordinating all the dynamic components of modern software systems is far more complex than simply writing code.
In project management, we call this phenomenon the "90% syndrome". (TechFlow note: In simple terms, the "90% syndrome" refers to the fact that a project is 90% complete, but the remaining 10% may require another 90% of the time and effort.)
The "90% syndrome" is the reason why Sam Altman's classic quote has become widely circulated:
As a result, many tech teams and managers will develop unrealistic expectations during the rapid iteration process, believing they can maintain high-speed progress indefinitely.
However, reality will always bring us back to our senses. When engineers and board members realize the law of diminishing returns, they have to slow down until they find a new breakthrough point and enter the next cycle.
What does this mean for crypto investors?
For crypto investors, this means that AGI (Artificial General Intelligence) will still be just a "buzzword" in the next few years. At the same time, many will use this concept to promote their "projects".
Swarms: Crypto's Response to AGI
In the intersection of crypto and AI, a brand new narrative is taking shape, centered on "agents" - especially "swarm intelligence" composed of a large number of agents.
Agent Swarms refer to multiple agents coordinated through a specific framework, who, through collective collaboration rather than relying solely on computing power, can complete complex tasks. This approach effectively solves the bottleneck problems of hardware and algorithms.
As Tom Shaughnessy mentioned in his article:
"A crypto-AI-based AGI alternative is quietly emerging.
We often take it for granted that OpenAI will be the ultimate winner.
After all, they have top talent (although some have left), powerful computing resources, leading model releases, and a strong focus on reasoning capabilities.
However, understanding the alternatives in this field is not always easy, as they are not always in the most obvious places.
The core of this alternative is hundreds of millions of narrow and highly specialized AI models (or agents). These agents are "experts" in their respective fields, and they do not need to reason comprehensively about everything, but form a "swarm intelligence" through collaborative work, which is far superior to a single large model. In fact, hundreds of millions of narrow models have always been the basis of my initial theory.
Developers can customize the reasoning paths of these agents (i.e., the agents' thought chains, such as when to stop researching or when to switch to a new direction), flexibly combine data and real-time information sources, use multiple base models (such as Nous Research, Prime Intellect, Llama, deepseek, or other open-source models), and deeply tune the details of each agent to make it fully focused on a specific task.
This "Cambrian explosion" of large-scale agent swarms is enabled by the funding of crypto tokens and driven by the decentralized crypto community. This model exhibits huge differentiation advantages in creating models and agents, which could never have been born in traditional Web2 AI labs. In comparison, its development speed and community support are unmatched.
Once we can access these "swarms" (i.e., the combination of expert models) through a simple and easy-to-use interface, and the "swarm" can intelligently select the most suitable models to execute tasks, this model will be widely accepted overnight.
The trend of technological development indicates that AGI is more likely to be built in an open form on decentralized blockchains, rather than being subject to potentially closed centralized platforms.
It's just a matter of time, and crypto-AI is becoming the leading path to collective AGI, with very promising prospects."
In fact, when we reach the limits of hardware performance, research progress, and physical laws, we always return to a familiar direction - aggregation.
Tom mentioned the term "Mixture of Experts (MoE)", but the concept is not that complex. Through agent swarms, blockchains have demonstrated their unique value in the field of artificial intelligence: coordination capability.
Thanks to the unique advantages of encryption technology in large-scale behavioral programming, we are able to deploy and manage a large number of intelligent agents globally. This allows us to build smaller, more focused large language model networks, where these models compete with each other and strive to provide the best service to end-users.
Last July, we discussed this in detail on the Club Cod3x podcast:
If not AGI, what will the future be?
As the crypto and AI fields continue to mature, we will make significant progress in curation, distribution, and commercialization. Although Web3 AI companies are still in the early stages, the potential of this field has already generated widespread attention.
I have been developing in the field of artificial intelligence combined with cryptocurrencies for several years. During this period, I have summarized some experiences about what methods are effective and what methods are ineffective.
Here is my latest analysis of the current artificial intelligence and cryptocurrency field:
1. Frameworks - Platforms for accelerating development, standardizing protocols, and facilitating communication between agents.
@virtuals_io - Social Framework (Virtuals)
@ai16zdao - Social Framework (G.A.M.E.)
@Cod3xOrg - Financial Framework (Moon)
@gizatechxyz - Financial Framework
@AlloraNetwork - Training Framework
@opentensor - Training Framework
@chirperai - Coordination Framework
@autonolas - Coordination Framework
