jakiro
Author: teng yan, former Delphi Digital NFT research supervisor; translation: 0xjs@Bit chain vision world
I am very fascinated by the idea of decentralized collective wisdom, so recently, I have been exploring the Bittensor model to guide the launch of AI networks in specific fields.
But I found that Bittersor had several serious defects.My discovery was shocking and made me pause.
Bittersor quickly review
Source: bittensor.com
Bittensor is a decentralized incentive network operation as an AI model.
Among its 32 subnets, the miners deploy AI models to compete according to specific standards.The verifications then evaluated and ranked these outputs, and the miners earned the TAO token based on their relative performance.At present, these subnets are dealing with tasks such as text generation, image generation, text conversion voice, and fine -tuning models.
“AI Model Horse”: Use the image generated by Dall-E 2
In short: This is the horse racing of the AI model.The best horse (model) gets the biggest reward.
The important idea here is that currency incentives naturally attract the best models and innovators to the most related subnets.
For the first time, I shared some of my early views on Bittersor in November last year.Since then, Bittersor has gradually become the darling of Crypto Twitter and the flag bearer of decentralized AI movement.As of now, Bittensor’s market value is US $ 2.8 billion, and the fully diluted valuation is US $ 8.5 billion.
The road to move forward is rugged
In my opinion, Bittersor’s biggest problem is:
1. The horizontal economy is broken
It is meaningful if it is implemented properly, and the competition of incentives in the subnet to produce the best output is meaningful.
Please pay attention to the thick body and capital “if”.
There are challenges such as inspirational games, relay mining, and authenticants’ understanding of the correct output.I believe these problems can be solved through sufficient research and experiments.Some subnets have taken measures or amend their standards to prevent the China relay from mining and other improper behaviors.
However, when the competition is expanding horizontally, the economic model begins to collapse.
Sub -nets must compete with each other to get greater inflation rewards.Once the dynamic TAO is achieved, this will become more obvious (the subnet will have its own dynamic token, token holders can affect the distribution of the reward between the subnet)
This leads to a problem:
How should TAO holders compare the economic value of different subnets to the Bittersor network?For example:
(1) Text to voice -to -voice model (subnet 3) and and
(2) Screening the visual model of the X -ray disease?
The value generated in each case is completely different.You can explain this problem in various ways, but the answer is never clear.Relying on market forces to determine the value contribution of each subnet is not the best choice for long -term strategic development.
This is like asking each AAPL shareholder to vote for the next product that Apple should develop.
2. The competitive challenge of Ziwang
I doubt whether the Bittersor subnet is competitive in the basic generation AI task (such as general images and text generation).Some open source projects have performed well in these fields, and these projects are usually promoted by reputation, research contributions and public services.
Take the powerful large language model (LLM) LLAMA-3 as an example of the open source tool released by Meta as an example.Although the development of LLAMA-3 has invested billions of dollars, it can still be used for free and has received strong support from the community.Since its release last week, it has downloaded more than 500,000 times on HuggingFace.
The benchmark comparison shows that open source LLMs such as LLAMA-3 are already the best among the same kind, and it can be said that it is comparable to the exclusive models such as the GPT-4 of OpenAI.
So far, I haven’t seen Bittersor subnet achieved the results or functions that surpass the existing open source model.I did not see evidence that their performance was better than the established benchmark.At some point, you must ask yourself if they will do this.
3. High operating cost of smart networks
The network relies on its native currency TAO to a large extent.Miners mainly join the network when they discover the opportunity of computing arbitrage, especially when mining rewards exceed the calculation cost.
TAO’s demand depends on::
(1) speculators (a lot of worms, we will not discuss here)
(2) Add more Bittersor subnet, or increase the needs of miners or verifications.However, this growth often weakens the enthusiasm of existing participants.Due to the pressure of the economic benefits of the miners/verifications, the decline in token prices will lead to a decline in the anti -personality of online activities.
Source: taostats.io
The operating cost of Bittersor is very high, and pays a 7,200 TAO inflation of 7,200 TAO every day.This is equivalent to about $ 3.2 million per day or $ 1.2 billion per year.
This raises a key question: Where does this fund come from?Mainly speculators.But Bittersor will generate $ 1.2 billion in one year?Its valuation may be far beyond its own value.
4. Transparency
Although Bittensor provides a certain degree of decentralization, it is worth noting that it is not open -this is a vital difference.Although the participation of Ziwang is open to anyone, the internal mechanism is not transparent, resulting in a limited understanding of users’ understanding of the following activities:
-
What are the miners doing?
-
What algorithms do they run?
-
How is the end result?
This lack of transparency may limit the diversity of applications using subnet output development, because many developers need to clearly understand the underlying reasoning process.
Some thinking
I don’t oppose Bittersor.In fact, I found that it is one of the most interesting experiments in the emerging Crypto X AI field.
It successfully gathered a group of smart people around its vision.It may stimulate many future projects.However, Bittersor must deal with the current challenges in order to have the opportunity to succeed in the real world, otherwise it will be lost in the grave of speculative hype.
Some end ideas:
-
If all subnets in the network are targeted at specific departments rather than scattered in various unrelated tasks, they can achieve a greater synergy.
-
For example, I imagine “Bittensor for genomic research/drug discovery” or “Bittersor for finance/transactions”.This can actually do well.This vision can promote economic cohesion, strategic direction, and business development efforts.
-
Bittensor’s economic model (in the current state) may be most suitable for the ecosystem of AI. These areas will not attract widespread attention, but they can benefit from targeted incentive measures.Potential applications include tasks specific to cryptocurrencies, such as predicting token prices.
Finally, I firmly believe that the decentralization of AI played an extremely important role in the world -the global dominance of large -scale technology companies and promoting innovation, especially through open source measures.I will explain my thoughts in detail in the future articles.
