AI requires ZK : ZK proofs for trust assurance of on-device execution LLMs The ZK Coprocessor project @lagrangedev made an announcement about providing ZK proofs for Google's Gemma3 at the "Verifying Intelligence" event hosted by Google Cloud and HouseofZK. Gemma 3 is a lightweight, high-performance "on-device" LLM developed by Google. Unlike large AI models like GPT that run on a central server and return results, Gemma3 runs and infers on the user's device as a lightweight model. On-device LLMs enable the use of AI without a network connection, ensuring privacy since data does not need to be sent to a central server. Additionally, since the majority of the costs associated with AI usage like GPT come from the computational expenses of the central server, this can significantly reduce those costs. However, having the model run locally also means that one cannot expect trust from a central server. There is a possibility that the local execution environment could be tampered with, and the inference results must be fully trusted 'by themselves.' In such cases, ZK coprocessors like Lagrange's DeepProve are used. DeepProve generates the output of Gemma3 executed locally in the form of ZK proofs. Through these ZK proofs, users of Gemma3 can automatically verify that - the model received the correct input, - Gemma3 computed correctly as designed, - without exposing sensitive information. In fact, the combination of this type of lightweight local execution LLM and ZK coprocessors aligns very well with blockchain + AI. It enables reliable distributed AI by providing inference results through lightweight high-performance LLMs like Gemma3 on decentralized blockchain nodes and submitting them as ZK proofs. Ultimately, in the era of on-device execution of AI, an infrastructure for ZK proofs that can guarantee trust is essential. AI-specific ZK coprocessors like Lagrange's DeepProve will play a crucial role in this.

@lagrangedev Be careful, it's not @Gem3a.