The underlying reason for Aethir's cooperation with Aphone: to make up for the lack of centralized computing power support

Original author: Haotian, crypto researcher

Editor's note: In April this year, Aethir announced a partnership with the decentralized cloud smartphone APhone. Aphone will make full use of localized edge computing capabilities to support the implementation and operation of applications across games, artificial intelligence, virtual reality/augmented reality, live streaming and other fields. On May 13, APhone stated on social platforms that it would airdrop to all Aethir node operators and node license owners. Crypto researcher Haotian analyzed the deep-seated reasons behind Aethir's series of measures. BlockBeats reproduced the full text as follows:

I was very curious, as a distributed GPU cloud computing provider, why did @AethirCloud launch the smartphone Aphone and airdrop it to Aethir node operators? In the past two days, I found the answer from everyone's heated discussion about GPT4o. The logic behind this is also related to the B side of OpenAI's "computing power explosion": the distributed computing power collaboration framework that is in urgent need of an explosion.

Take GPT-4o as an example. On the surface, it can allow large models to be connected to robots and other devices, so that human-computer interaction can achieve a "smart" cross-era novel experience. But it's okay if the prompt is text. To turn it into pictures, videos and other content, you need to have strong computing power to slice and frame the pictures, videos and other content first, and then convert them into semantics before interacting with the large model. If the content of the real-time interactive screen is too complex, the parsing and feature extraction process will generate a large number of tokens, and the corresponding computing power consumption will also be huge, so it is difficult to achieve no delay.

Therefore, the real-time, delay-free interaction achieved by companies such as OpenAI relying on centralized computing power cannot be regarded as flawless innovation, but rather more like a hastily handed in answer sheet. If you look closely, there are a lot of problems.

What should I do? The solution to this problem still needs to rely on the rise of the distributed cloud computing service market. Users’ terminal devices, such as mobile phones or computers, can perform model analysis and pre-processing, rely on edge computing to complete some real-time input analysis, and then transmit it to the cloud to interact with the large model. In theory, this seems to reduce the reliance on centralized computing power, optimize latency, and make up for the shortcomings of large model evolution driven by pure centralized computing power.

This may be the potential reason why Aethir launched the Aphone smartphone: let smart hardware as a terminal carrier also become part of the distributed cloud computing processing center, and then refine the application scenarios of model training to make up for the shortcomings of relying solely on centralized computing power.

A distributed cloud computing power collaboration framework can: 1. Aggregate a large amount of distributed computing power and data resources, and use edge computing capabilities to reduce dependence on purely centralized computing power; 2. Reduce dependence on a single service provider and improve the fault tolerance and stability of distributed systems; 3. Drive interoperability between different models and services, and enrich the service scope and experience of large language models.
Note: Recently, I have focused on the AI track. For market opportunities, development prospects, and project representative system introductions of decentralized computing power services, please refer to @web3caff_zh's 10,000-word research report for additional reading: https://research.web3caff.com/zh/archives/17351?ref=743

Original link

欢迎加入律动 BlockBeats 官方社群：

Telegram 订阅群： https://t.me/theblockbeats

Telegram 交流群： https://t.me/BlockBeats_App

Twitter 官方账号： https://twitter.com/BlockBeatsAsia