This is part of our series on a year of Bittensor experience, leading up to our anniversary, tomorrow, the 13th of July. After discussing miner archetypes, incentive landscapes, validator technicalities, a new validator design and concept, suitable subnet subjects are discussed. Please let us know what you think in our Discord channel!
We have seen a lot of subnets and subnet ideas, touching on many subjects. Sometimes the goal of a subnet is very clear, sometimes you can’t make any sense of it, no matter how often you read the whitepaper or ask for clarification.
Monetization and product focus seem to be the required features for all subnets, post-dtao. A fundamental issue with that, is that in many cases, performance on real-world tasks or challenges is hard to measure. To mitigate this, “synthetic” challenges are mixed into the set of “organic” challenges given to miners, so that the validator “knows the right answer” and can score the miner purely based on synthetic challenges. Ideally, the miner wouldn’t know which category a challenge belongs to – in practice miners can often tell them apart easily.
The optimizing miner of course tries to find the best scoring answer, using whatever method possible – and as we explained, there are many – and the subnet may diverge wildly on performance, when comparing synthetic and organic response quality. Or to put it in simple terms: the subnet doesn’t deliver on its promise in the real-world application it was built for.
Subnets that fall between research and real-world applications may start off as promising, but often see their emission drop after a few weeks. At first they seem to deliver on their initial promise, but then it turns out that either the subnet is meta-gamed to death, or that the challenge is basically solved, turning the subnet into a boring tao distribution machine.
Subnet owners often struggle (and fail) to stay ahead in this game. Not due to lack of skill on their end, but because the subnet concept is too fragile and not fully crystallized yet, and the miners are simply too good. It’s an uphill battle for a small subnet team against a large and diverse group of miners. Remember the many miner types: if the overanalyzing miner can’t poke a hole in the concept, maybe the blockchain-savvy miner can, or the calculative, rogue or devops miner, for that matter.
Another way to turn a subnet into a boring tao distribution machine, is to require no intellectual or creative effort whatsoever from miners. Prime examples of this are subnets that merely resell digital goods or services. Yes, they tick the box of “revenue generation” – but that’s about it. They often also fail to demonstrate how value is added in the economic chain – revenue yes, profit no. Normally you would expect subnet owner, validator and miner to each contribute value to the final product. If the subnet only resells something, where the retail price is nearly identical to the wholesale price, the economics don’t check out, and the concept is not a long-term fit for Bittensor. In other words: there should be a significant gross profit margin.
Summing up: A good dtao-era subnet should solve a real-world problem, requiring intellectual or creative effort put in by miners, with a significant gross profit margin, that scales “infinitely” in volume and complexity, and doesn’t rely on a limited set of synthetic queries to assess miner quality. Monetization and product focus should be clear from the start, not something left to figure out later.
This is the last article in our series on a year of Bittensor experience. Tomorrow is our anniversary where we will unfold our plan for the coming year. And change the font.
0 Comments