Evaluating Large Language Models Trained on Code
At a glance
Introduced OpenAI Codex and HumanEval, a benchmark that tests whether AI-generated Python code works.
In one paragraph
OpenAI introduced Codex, a language model trained to write code from natural-language instructions. The paper also released HumanEval, a benchmark that runs tests to check whether generated Python programs work as intended.
Authors 58
Organizations
Credited for this work at publication.