Training a top-tier large model currently consumes enough energy to power a small city for several days. The deeper issue is that 78% of the global AI Computing Power is tightly held by a few tech giants, and the rental costs for high-end GPUs are exorbitantly high. The result is that these giants hoard computing power, competing with each other while possibly doing redundant tasks, leading to diminishing marginal returns; meanwhile, countless small and medium-sized development teams, research institutions, and innovative startups are blocked from innovation by the high costs of computing power. This is not just an economic issue; it is a significant misallocation of resources that is completely contrary to the goals of sustainable development.
So some have proposed a new approach. Instead of directly producing Computing Power, it is better to construct a decentralized global Computing Power resource "smart grid" through clever economic design and protocol architecture. The goal is clear: to ensure that every joule of computational energy flows to the places that need it the most and can create the most value.
The key is how to incentivize. Traditional blockchain's PoW has been criticized for its energy consumption for many years, while PoS primarily rewards holders. This new consensus mechanism has completely changed the thinking - it's called "Attribution-based Intelligent Proof," abbreviated as PoAI. It does not reward meaningless hash calculations, nor does it reward merely holding coins, but rather rewards those participants who truly create actual, verifiable value for the AI network. For example, nodes that provide high-quality Computing Power and successfully complete valuable AI tasks.
In this way, when a developer wants to train a model, he can find the computing power resources he needs in a decentralized network. In this model, the income of computing power providers is directly linked to their actual contributions, and the incentive mechanism is clear. In simple terms, it activates idle computing power resources, allowing innovators to acquire what they need at a reasonable cost, breaking the monopoly of giants while also improving the overall resource utilization efficiency of the ecosystem.
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
13 Likes
Reward
13
3
Repost
Share
Comment
0/400
gas_fee_therapy
· 12-23 15:37
The idea of PoAI is quite interesting, and it is indeed an improvement compared to the purely energy-consuming operations of PoW.
View OriginalReply0
SchrodingersPaper
· 12-23 15:35
PoAI? Sounds like another savior solution, but I can't shake the feeling that it'll still be played for suckers by large investors...
Forget it, I'll just stock up on some spare GPUs, who knows when they might be worth something.
This time is really different, the decentralized computing power grid is the future direction, I believe it.
78% is held by giants? Serves them right, if retail investors had a bit of sense, they should get in and become nodes...
Wait a minute, how does this thing verify "actual value"? Isn't it still dark box operations?
I've heard too many times about breaking monopolies, and in the end, it just becomes a new monopoly, with periodic plays for suckers.
PoAI is much more ethical than PoW, truly valuing output instead of wasting on electricity, I need to study the White Paper.
View OriginalReply0
fren.eth
· 12-23 15:25
This PoAI sounds good, but how do we actually verify "verifiable value"? Is it just another trap?
Training a top-tier large model currently consumes enough energy to power a small city for several days. The deeper issue is that 78% of the global AI Computing Power is tightly held by a few tech giants, and the rental costs for high-end GPUs are exorbitantly high. The result is that these giants hoard computing power, competing with each other while possibly doing redundant tasks, leading to diminishing marginal returns; meanwhile, countless small and medium-sized development teams, research institutions, and innovative startups are blocked from innovation by the high costs of computing power. This is not just an economic issue; it is a significant misallocation of resources that is completely contrary to the goals of sustainable development.
So some have proposed a new approach. Instead of directly producing Computing Power, it is better to construct a decentralized global Computing Power resource "smart grid" through clever economic design and protocol architecture. The goal is clear: to ensure that every joule of computational energy flows to the places that need it the most and can create the most value.
The key is how to incentivize. Traditional blockchain's PoW has been criticized for its energy consumption for many years, while PoS primarily rewards holders. This new consensus mechanism has completely changed the thinking - it's called "Attribution-based Intelligent Proof," abbreviated as PoAI. It does not reward meaningless hash calculations, nor does it reward merely holding coins, but rather rewards those participants who truly create actual, verifiable value for the AI network. For example, nodes that provide high-quality Computing Power and successfully complete valuable AI tasks.
In this way, when a developer wants to train a model, he can find the computing power resources he needs in a decentralized network. In this model, the income of computing power providers is directly linked to their actual contributions, and the incentive mechanism is clear. In simple terms, it activates idle computing power resources, allowing innovators to acquire what they need at a reasonable cost, breaking the monopoly of giants while also improving the overall resource utilization efficiency of the ecosystem.