We've just completed a full, verifiable AI inference pipeline: - Execute the model locally on Apple Metal - Generate cryptographic proofs of the computation process - Perform on-chain streaming verification on @Starknet This is more than just "AI generating an output." It provides a path to verifiable AI systems. For this run, we used @Alibaba_Qwen Qwen2-0.5B, executed it locally, captured the forward propagation process, and verified it through a 6-step on-chain process. This means that runtime, commitment, proof, and blockchain verification all work together in a real-time pipeline. What impressed us most was that this didn't just happen in a dedicated data center environment. Our ability to run it locally on Metal is crucial because the future of verifiable AI is likely to extend beyond hyperscale data centers. It must scale to local devices, edge computing, enterprise systems, cloud GPUs, and ultimately, decentralized computing networks. The core idea here is completeness. It's not just about speed, cost, or latency; more importantly, it's about being able to verify which model is running, whether the weights are correctly bound, and whether the output matches the executed computation path. In many cases, a "trusted provider" suffices. But in others, such as enterprise workflows, proxy systems, third-party inference, and auditable environments, stronger assurances become crucial!