NELA – Net-Based Executable Logic Automaton (github.com)

NELA (Net-Based Executable Logic Automaton) has been introduced as a new programming language specifically crafted for Large Language Models (LLMs), moving away from human-centric syntax to enhance machine comprehensibility. By encoding computation as a formally verifiable interaction net, NELA promises greater reliability in machine reasoning and logic compared to traditional text-based languages. The design renounces human readability but offers strong formal guarantees of stability, which is critical as AI systems increasingly manage essential infrastructure. This innovation is significant for the AI/ML community as it addresses the urgent need for formally verified code to ensure security and correctness in AI applications. NELA employs Interaction Net theory to achieve properties like strong confluence and linearity, while also integrating concepts from linear logic and dependent type theory, where programs can serve as proofs. The language consists of a surface syntax (NELA-S) for writing programs and a backend compiler (NELA-C) that transforms these into interaction nets for formal verification. Demonstrations through classic algorithms, such as quicksort and a stack-based virtual machine, underscore the language's capabilities and its potential to revolutionize how LLMs are deployed in programming.