爱范儿 · AI 筛选· 张子豪·· 5 hours agoSelectedAI score71
Anthropic Releases Open-Source Research Prototype FlashProve: Verifying True LLM Weights with Zero-Knowledge Proofs
Anthropic微微发力，AI中转站的天就要塌了？
AI guide
Anthropic has released FlashProve, an open-source research prototype that leverages zkSNARK zero-knowledge proofs to generate cryptographic fingerprints and proofs for LLM inference, allowing users to verify that outputs were computed by the agreed-upon model without accessing its weights.
Why it matters
The original post explains the mechanics and performance overhead of verifying genuine LLM weights via zero-knowledge proofs, offering a technical reference for detecting whether third-party APIs have quietly swapped models.
Source:爱范儿 · AI 筛选 · ifanr.com