FYLs-G2P is a hybrid G2P engine that bridges the gap between massive end-to-end models and simple dictionary lookups. It uses a contextual POS tagger, neural OOV inference, and weighted graph mapping to accurately pronounce words, including unseen words and homographs. The engine is designed for edge devices and real-time TTS, with a focus on robustness and morphological intelligence.

Anthropic researcher Nicholas Carlini used 16 instances of the company’s Claude Opus 4.6 AI model to build a C compiler from scratch. The AI agents, working independently with minimal supervision, produced a 100,000-line Rust-based compiler capable of building a bootable Linux kernel. While the project demonstrates the potential of AI coding agents, it also highlights their limitations and the significant human effort required to support their work.