Tag
This paper introduces a contract-grade verifier of twelve adversarial gates for checking LLM-generated GPU kernels, finding that 39.5% of kernels accepted by standard loose tests are broken. It also presents the first native Blackwell training backward kernel for the GDN (gated-linear-recurrence) family.
Kuna is an experimental decompiler whose code was almost entirely written by an LLM, achieving control flow structuring performance competitive with IDA Pro 9.2 through autonomous refinement based on scientific metrics.
A Rust implementation of the RAR compression format, written mostly by AI language models (OpenAI Codex and Claude), which would have taken years to develop manually but was completed in weeks at low cost.