Author: Minta
1/n (continuously updated) Start a post to record the secondary ideas of Crypto & AI. The Web3 & AI Sector layering is similar to Web2. The data layer/computing power layer is the bottom-level infrastructure, followed by the model layer, service layer/Agent layer, and finally the application layer.
2/n Narrative From a narrative perspective, the lower the degree of standardization/homogenization of a track, the greater the probability of Alpha. For example, in the computing power layer, GPUs are homogeneous, so computing power projects are mainly involved in rolling gameplay/GTM/computing power asset derivatives, etc. The lowest standardized sub-tracks are: model layer, data layer, and agent layer.
3/n Model layer The model layer is a track with very large variables. The emergence of a new model can quickly change the market landscape. For example, OpenAI's GPT-4o has brought unlimited possibilities for new human-computer interaction methods. Phi-3 released by Microsoft and OpenELM released by Apple on Hugging Face have rapidly advanced the process of mobile training. From a secondary perspective, large variables = large opportunities for non-consensus = large opportunities for Alpha.
4/n Model layer A good model platform has at least the following characteristics: 1. Model composability: support compatibility with multiple large models in rotation, rather than a single large model; 2. Need to understand/be compatible with the business logic of computing resources; 3. Supervised calculation/model scoring In summary, a model platform that supports calling the hybrid model library and Tools API and is compatible with computing resources is a model platform that can survive bull and bear markets.
5/n Bittensor Case The current top 1 in the model layer is undoubtedly Bittensor. In order to build an incentive layer, what did Bittensor do right? Let's briefly review the timeline of Bittensor (incomplete statistics): 1. Before October 2024: the community was established during the test network; 2. In October 2024, subnet registration was gradually opened: from the first batch of 9 subnets, to 32 subnets; in May 2024, it began to gradually expand to 64 subnets;
6/n Bittensor Case 3. Adjust the incentive model according to the ecosystem: from the universality of incentives to the survival of the fittest through incentives, for example: (1) The halving time is brought forward, and the first halving is adjusted from 25 years to 23 years; (2) Tokenomics adjustment, the direct incentive allocation ratio of the subnet is completely left to the market to decide, adding more dimensions of game; 4. Slowly forming a flywheel effect
7/n Bittensor Case From a secondary perspective, the "wealth-making effect" successfully created by Bittensor is very critical. Several decision-making turning points of Bittensor all stepped on the time node of liquidity surge. From September to November 23, the number of subnets was expanded successively to provide a large number of buying orders; at the same time, combined with its special staking mechanism, it quickly promoted the rise of token prices, thereby attracting more projects to register as subnets, forming a virtuous circle. Therefore, if you look for the sword in the boat, you will find that the subnet registration fee is the best vane of $TAO price, and the subnet registration peak and price peak always appear one after another.
8/n Bittensor Case As of the time of writing, the top 8 Bittensor Staking situations are shown in the figure. The top 8 Stakers have a combined Staking Rate of 61.2%, and the total Staking Rate of the project is 84.61%. The current MC is 2.5B, and it is estimated that MC can reach 10B+. You only need to predict the Staking situation of large users to complete the price estimation.
9/n Bittensor Competitors For Bittensor competitors of the same type, they need to solve the problem of high entry barriers. At present, the registration fee for TAO's subnet is still not low; and they need to rely on strong BD capabilities to roll out a large number of models in a short period of time and do a good job of market value management at the same time. Some early projects such as
@communeaidotorg, @zero1_labs, @arbius_ai and others are doing similar things. If their ecosystem expands rapidly, it may be a good entry point.
10/n Bittensor competitors Take Commune as an example. This is a startup project of TAO ecosystem core Builder. Compared with TAO: 1. Commune has a lower threshold and it is easier for Dev to register modules; 2. Commune’s incentive system is greatly simplified/deleted, and decisions rely on a simpler voting system; In general, there are currently no projects in the secondary blockchain that can compete with Bittensor in the short term.
11/n Model layer un-coined project alternatives @Nimble_Network built a global orchestration layer to achieve general AI operations and full-link access; @Gatling_X launched an EVM that supports computing scenarios; @ritualnet takes a multi-pronged approach, from incentive networks, to linking distributed computing devices, model hosting, sharing, reasoning, optimization, etc., to the API layer for accessing models, as well as anti-censorship and privacy.
