Show HN: Glyph a blockchain where the proof-of-work is neural network inference (github.com)

A new blockchain project named Glyph has been unveiled, where the proof-of-work mechanism is based on neural network inference rather than traditional computational methods. Developed by Claude Fable, this innovative approach requires miners to run a pinned open-weights transformer model on salted prompts. The model's output is compressed into a unique fingerprint using a specific algorithm and then hashed to meet a difficulty target, similar to Bitcoin. Notably, Glyph's integer-only inference engine, introduced in its latest protocol version, ensures that the mining process is deterministic across different hardware platforms, eliminating discrepancies seen in earlier models. This development is significant for the AI/ML community as it merges machine learning functionality with blockchain technology, creating a novel paradigm where AI inference is a core component of the mining process. The validation process involves rigorous cross-hardware testing, demonstrating robust stability and security through consensus mechanisms. With challenges like adversarial attacks and model vetting embedded into the framework, Glyph also emphasizes the importance of securely integrating AI models into its ecosystem, making it potentially a game-changer for decentralized AI applications.