CodeNet

A new arXiv paper (2609.16454v1) finds that supervised fine-tuning (SFT) is not inherently biased toward mode collapse or its opposite, over-dispersion, in large language models. The authors derive a …

A 2026 journal study found that trillion-parameter large language models (LLMs) such as GPT-4, LLaMA, and Falcon are capable of literate programming, while smaller models excel at recognizing programm…

IBM open-sourced CodeAlchemy, a synthetic dataset of nearly 1 trillion tokens across 15 programming languages, which may be the largest synthetic code dataset to date. The dataset includes 1.3 million…