jakiro
Author: 0xai, AI creation platform Source: Medium Translation: Shan Oba, Bitchain Vision Realm
What is Bittersor?
Bittersor itself is not an artificial intelligence product, nor does it produce or provides any artificial intelligence products or services.Bittersor is an economic system that provides AI product producers with a very competitive incentive system and acts as optimizers in the AI product market.In the Bittensor ecosystem, high -quality producers have obtained more incentives, and producers with poor competitiveness have gradually been eliminated.
>
0xai team, Jacob and Vitalik in the group discussion
So, how is Bittensor specifically to create an incentive mechanism for encouraging effective competition and promoting the organic production of high -quality AI products?
>
Bittensor flywheel model
Bittersor achieved this goal through the flywheel model.The verifications evaluate the quality of artificial intelligence products in the ecosystem, and incentives according to their quality distribution to ensure that high -quality producers get more incentives.This stimulates the increasing increase in high -quality output, thereby increasing the value of the Bittersor network and increasing the appreciation of TAO.The appreciation of TAO not only attracted more high -quality producers to join the Bittersor ecology, but also increased the attack cost of the manipulator’s manipulation of quality evaluation.This further strengthened the consensus of honest validations, enhanced the objectivity and fairness of the evaluation results, and achieved more effective competition and incentive mechanisms.
Ensuring the fairness and objectivity of the evaluation results are a key step for rotating the flywheel.This is also the core technology of Bittersor, that is, an abstract verification system based on Yuma consensus.
So, what is the Jade Horse Consensus, how can it ensure that the quality evaluation after consensus is fair and objective?
Yuma consensus is a consensus mechanism that is designed to calculate the final assessment results from the diverse evaluation provided by many verifications.Similar to Byzantine’s fault consensus mechanism, as long as most of the authenticants in the network are honest, they can eventually reach the correct decision.Assuming that honest verifications can provide objective evaluation, the results of the evaluation after consensus will also be fair and objective.
Taking the subnet quality assessment as an example, the root network verification device evaluates and rank the quality of each subnet output.Summarize the evaluation results of 64 validator and obtain the final evaluation results through the Yuma consensus algorithm.Then use the final result to allocate the newly created TAO to each subnet.
At present, Yuma’s consensus does still have room for improvement:
-
Root network validators may not completely represent all TAO holders, and the evaluation results they provide may not necessarily reflect a wide range of views.In addition, the assessment of some top -level verifications may not always be objective.Even if the prejudice is found, it may not be corrected immediately.
-
The existence of root network verifications restricts the number of subnets that can be accommodated by Bittensor.It is not enough to compete with centralized artificial intelligence giants. It is not enough to have only 32 sub -nets.However, even if there are 32 subnets, the root network verification device may be difficult to effectively monitor all subnets.
-
Verivers may not have a strong willingness to migrate to the new subnet.In the short term, the verificationrs migrate from the high emissions from the old sub -net to the Xinzi.com with a lower emissions.Whether the emission of Xinzi.com can eventually catch up, coupled with the clear losses of the reward in the process of pursuit, weaken their willingness to migrate.
-
The dynamic TAO scattered the power of the subnet quality to all TAO holders, not a small number of verifications.TAO holders will be able to indirectly determine the distribution ratio of each subnet through pledge.
-
Without the restrictions of root network validators, the maximum number of active subnets will increase to 1024.This will greatly reduce the threshold for new teams to join the Bittensor ecosystem, leading to more intense competition between subnets.
-
Earlier authenticants migrated to Xinzi.com may get higher rewards.Moving to the new subnet as soon as possible means to purchase the DTAO of the subnet at a lower price, thereby increasing the possibility of receiving more Tao in the future.
-
Due to similar positioning of subnets, resource redundancy and duplicate.Of the 32 Subnet, multiple SUBNET focuses on popular directions such as text transfer images, text prompts, and price prediction.
-
There are sub -nets without actual use cases.Although the price prediction subnet as a prophet provider may have theoretical value, the current performance of the predicted data is far from being used by end users.
-
Example of “inferior currency expels good currency”.Some top -level verifications may not have a strong willingness to migrate to the new subnet, even if some new subnets show significantly higher quality.However, due to lack of financial support, sufficient emissions may not be obtained in the short term.Because the protection period after Xinzi Online is only 7 days, if you cannot quickly accumulate sufficient emissions, you may face the risk of being eliminated and offline.
Bittersor also plans to upgrade mechanisms to solve these shortcomings:
Strong tolerance is also a major advantage of Yuma consensus.YUMA consensus is not only used to determine the emissions of each subnet, but also to determine the distribution ratio of each miner and verifier in the same subnet.Moreover, no matter what the miner’s task is, the contribution it contains, including computing power, data, human contribution and intelligence, is abstract consideration.Therefore, the Bittensor ecosystem can be connected at any stage of AI commodities to enjoy the value of the Bittersor network while enjoying the incentives.
Next, let’s explore some leading subnets and observe how Bittersor inspires the output of these subnets.
Sub -network
Ziyang 3: MyShell TTS
You can contribute to the development of MyShell AI/MyShell TTS subnet by creating an account on GitHub.
Circulation: 3.46% (April 9, 2024)
background: MyShell is behind the team behind myshell TTS. The core members come from well -known universities such as Massachusetts Institute of Technology, Oxford University, Princeton University.MyShell aims to create an unclear platform that allows college students who do not program background to easily create the robot they want.MyShell focuses on the TTS field, audiobooks and virtual assistants. In March 2023, the first voice chat robot Samantha was launched.With the continuous expansion of the product matrix, the current registered users have exceeded one million.The platform custody various types of robots, including language learning, educational and practical robots.
Positioning: MyShell launched this Subnet to gather the wisdom of the entire open source community and create the best open source TTS model.In other words, myshell TTS does not directly run the model or process the request of the end user; on the contrary, it is a network for training TTS models.
>
MyShell TSS architecture
The process of myshell TTS running is as shown in the figure above.Miners are responsible for training models and upload the training model to the model pool (the metadata of the model also stored in the BitTensor blockchain network); the verifier is evaluated by generating test cases, evaluating the performance of the model and scoring according to the results; the Bittersor areaThe blockchain is responsible for using Yuma consensus aggregation weight to determine the final weight and distribution ratio of each miner.
In short, miners must continue to submit higher -quality models to maintain their rewards.
At present, myshell has also launched Demo on its platform for users to try the model in MyShell TTS.
>
>
In the future, as the model of MyShell TTS training becomes more reliable, more use cases will be launched.Moreover, as an open source mode, they will not only be limited to myshell, but also expand to other platforms.Training and inspiring open source models through this decentralization method is not exactly our goal in decentralized artificial intelligence?
Ziwang 5: Open Kaito
You can contribute to the development of Open Kaito by creating an account on GitHub.
Issue: 4.39% (April 9, 2024)
Background: The team behind Kaito.ai is the Open Kaito team. Its core members have rich experience in the field of artificial intelligence. They previously worked in first -class companies such as AWS, Meta and Citadel.Before entering Bittensor Ziwang, they launched the flagship product Kaito.ai -a web3 chain data search engine, launched in the fourth quarter of 2023.Using artificial intelligence algorithms, Kaito.ai optimizes the core components of search engines, including data collection, ranking algorithms and retrieval algorithms.It has been recognized as a first -class information collection tool in the encrypted community.
Positioning: Open Kaito aims to establish a decentralized index layer to support intelligent search and analysis.Search engine is not just a database or ranking algorithm, but a complex system.In addition, an effective search engine requires low latency, which puts an additional challenge to the construction of a decentralized version.Fortunately, these challenges are expected to be resolved through Bittensor’s incentive system.
>
The operating process of Open Kaito is shown in the figure above.Open Kaito does not simply disperse each component of the search engine, but defines the index problem as a miner verification problem.In other words, the miners are responsible for responding to the user’s index request, while the verifier distributes the demand and scores the miner’s response.
Open Kaito does not limit how the miners complete the index task, but pay attention to the final output of the miners to encourage innovative solutions.This helps create a healthy competitive environment between miners.Faced with user indexes, miners strive to improve their implementation plans to obtain higher quality response results with less resources.
Ziwang 6: Nous Finetuning
You can contribute to the development of Nous Research/Finetuning subnet by creating an account on GitHub.
Issue: 6.26% (April 9, 2024)
Background: The team behind Nous Finetuning comes from Nous Research, a research team that focuses on large -scale language models (LLM) architecture, data synthesis, and in -equipment in -equipment reasoning.Its co -founder was the chief engineer of Eden Network.
Positioning: Nous FineTuning is a subnet dedicated to fine -tuning large language models.In addition, the data used for fine -tuning also comes from the Bittensor ecosystem, which is specifically the subnet 18.
Nous FineTuning’s running process is similar to MyShell TSS.Miners are based on the data training model from Zhewang 18, and regularly release these models to host on the Hugging Face; the verifier evaluates the model and provides scores; similarly, the Bittensor blockchain is responsible for using YUMA consensus to aggregate the weight to determineThe final weight and circulation.
Ziwang 18: Cortex.t
Corcel-API/Cortex.t can be contributed by creating an account on github.
Issue: 7.74%(April 9, 2024)
Background: The team behind CORTEX.T is Corcel.io, which has received support from MOG, the second largest verification of the Bittersor network.Corcel.io is an application for end users. It provides a similar experience to the ChatGPT by using the Bittensor ecosystem’s artificial intelligence products.
Positioning: Cortex.t is positioned as the last layer before the result of the result with the end user.It is responsible for detecting and optimizing the output of various seed networks to ensure that the results are accurate and reliable, especially when a single prompt calls multiple models.Cortex.t aims to prevent blank or inconsistent output and ensure seamless user experience.
Miners in Cortex.t use other subnets in the Bittensor ecosystem to process the request of the end user.They also use GPT 3.5 Turbo or GPT 4 to verify the output results to ensure the reliability of end users.The verifications evaluate the output of miners by comparing the results generated by Openai.
Ziwang 19: VISION
Contributing Namoray/Vision’s development by creating an account on GitHub.
Issue: 9.47%(April 9, 2024)
Background: The development team behind Vision also comes from Corcel.io.
Positioning: VISION aims to maximize the output capacity of the BitTensor network by using a optimized subnet called DSIS (distributed scale to modify subnet).The framework accelerates the response of the miners to the verification.At present, Vision focuses on the scene of image generation.
Verifications receive demand from the front end of the Corcel.io and distribute it to the miners.Miners can freely choose their favorite technology stacks (not limited to models) to deal with demand and generate response.Then, the verifications evaluate the performance of the miner.Due to the existence of DSIS, Vision can respond faster and more effectively than other subnets.
Summarize
From the above example, it can be seen that Bittensor shows a high degree of tolerance.Miners’ generation and verification verification occurred under the chain, and the Bittersor network only distributed rewards to each miner based on the evaluation of the verification.Any aspect of artificial intelligence products suitable for miners’ verification device architecture can be converted into subnets.
Theoretically, competition between subnets should be fierce.To continue to get rewards for any subnet, it must continue to generate high -quality output.Otherwise, if the root network verification device believes that the output value of a subnet is low, its distribution may be reduced and eventually replaced by new subnets.
But in reality, we did find some problems:
These issues reflect that the competition between subnets is insufficient, and some verifications have not played a role in encouraging effective competition.
The Open Tensor Foundation Verification (OTF) has implemented some temporary measures to alleviate this.As the biggest verification owner with 23% of the pledge right (including commission), OTF provides a channel for Ziwang to compete for more Staked Tao: Ziwang owner can submit requests from OTF every week to adjust its Staked Tao in the subnet.Proportion.These requests must cover 10 aspects such as “subnet targets and contribution to Bittersor ecosystem”, “subnet reward mechanism”, “communication protocol design”, “data source and security”, “calculation requirements” and “roadmap”, To facilitate OTF’s final decision.
However, to fundamentally solve this problem, on the one hand, we urgently need to launch DTAO (Dynamic Tao), which aims to fundamentally change the above unreasonable problems.Alternatively, we can call on large -scale verificationrs holding a large number of STAKE TAO more from the perspective of “ecosystem development” instead of just considering the long -term development of the Bittersor ecosystem from the perspective of “financial returns”.
In summary, with its strong tolerance, fierce competitive environment and effective incentive mechanism, we believe that the Bittersor ecosystem can organically produce high -quality artificial intelligence products.Although not all the output of the existing subnet can be comparable to the output of centralized products, we should not forget that the current Bittersor architecture has just been established an anniversary (Ziwang #1 registered on April 13, 2023).For a platform with potential and centralized artificial intelligence giants, we may focus on proposing a practical improvement plan, rather than eager to criticize its shortcomings.After all, we don’t want to see artificial intelligence constantly controlled by a few giants.
