Remember when I told you that storing ZK proofs on Arweave would be the next big use case? Now it's coming to fruition through Arweave's partnership with @inference_labs. The second biggest Bittensor subnet and the biggest producers of ZK machine learning proofs. The problem with verifying the activity of AI agents is that you would normally need the inputs and outputs to verify the correctness of the result. But users don't want their inputs exposed; imagine if every LLM prompt you ever wrote was stored publicly on the blockchain (I shiver at the thought..) That's why @inference_labs created their Omron zkML protocol. It works as follows: - The user sends their inputs to the miner who picks a model and computes an output and a corresponding ZK proof. - Only the output and the proofs are inscribed to the blockchain. - Then a validator verifies the proof, rates the miner on accuracy and latency, and returns the result to the user if valid. But for the system to be trustless, you need these proofs and outputs to be accessible always and to anyone to verify, and this is not a small amount of information. Think about long text responses from LLMs, images generated by diffusion models, and now even videos. You need somewhere trustless, cheap, and publicly accessible to store them along with the corresponding proofs. That's where the Arweave partnership intervenes. 🧵(1/?)