Sentiment detected Tao last week (again), and this time looking into it with great depth. Bittensor is complicated - and still grasping all sides. But I have put below how I am thinking about it is so far (it may change the more I read and as I look through more code) 1. It’s a simple core blockchain itself - that gets validated and blocks produced. This only provides consensus and rewards. 2. The rewards aren’t to run the blockchain - but incentives the utilization of it. The blockchain emits rewards via Tao into areas where it is used. These use cases are called subnets. Currently only technical subnets exist (currently about 100) AI infrastructure, storage, image classification, and then some data analytics. 3. Each subnet is its own world - and anything can happen in it. The blockchain doesn’t care about what happens on a subnet and data isn’t computed onto it. There is no impact of scaling due to activity in a subnet. 4. There are validators and miners within each subnet - they perform the work and then agree what was correct. They use mostly use AI to agree on quality. For example was the LLM response correct or not for the users query? The miners and validators test each other by also working on AI to AI generated sample user content feed with real user content so the two are indistinguishable. This allows them to work and fine tune themselves faster. 5. Only the top 64 validators earn reward tokens in a subnet. This forces a race of best results in each subnet. Similarly to the subnets themselves the more stake of Tao from delegators (users) the more rewards they’ll receive. 6. Unlike Solana you can gain and lose by staking. You stake your Tao into a subnet- and that converts your Tao into what is called an alpha token. The exchange between the two is handled by a AMM (automated market maker) contract with a two sided ratio of Tao in the pool to alpha token left. Same as how all the DEXs currently work. The more people stake into the subnet the more expensive the conversion is from Tao to alpha. You have a first in advantage same as other tokens. 7. By staking you earn a proportional distribution of 41% of emissions (paid in alpha - the subnet token) and rewards (paid in Tao to the pool) of the subnet. 8. Creating a subnet yesterday would have cost 355 Tao so around $100k - only one can be created per day. The direct reward for a subnet is the 18% distribution of the rewards and emissions. 9. Owning subnet is likely to be by teams that will use the underlying output of the work of the subnet. For example if you had an end user application like a FaceSwap tool. Reasons the team would use a Bittensor subnet. - don’t want to have one large model providing all the results of the face swapped images - don’t want to manage the infrastructure and scaling issues related. - crowdsource updates to the models and improvements in quality to the FaceSwap output. - focus on the user application only and allow others to be incentivized for the models. The LLM models and inference would be done via the subnet in defi-ai manor (ai generated output - text, video and audio). The miners (LLM or GPU providers) would run the models and compete for the most cost efficient way to provide FaceSwap the output. Remember only top 64 get rewards. There is a lot more going on and will continue reading. One thing I have found is there is an EVM layer to it - which allows transfers, smart contracts but haven’t seen anyone talk about this yet. Comment questions or corrections - and I’ll write another summary.