Named Codex, reports suggest that the system is designed to not just be of use to trained full-time programmers, but will also be of assistance to amateur coders.
“Codex can now interpret simple commands in natural language and execute them on the user’s behalf—making it possible to build a natural language interface to existing applications,” says OpenAI in a blog post as it launched a private beta.
- Here’s our roundup of the best laptops for programming
- Get started with programming with these best Python courses
- Take a look at out collection of the best HTML courses
Codex is a descented of GPT-3, the third generation of OpenAI’s Generative Pre-trained Transformer, which was conceptualized to perform natural language processing (NLP) tasks.
Not quite there yet
According to OpenAI, Codex is trained on several billion lines of public code. The system is reportedly proficient enough to adapt to an individual developer’s coding style.
Notably, Codex powers GitHub Copilot, a new tool launched in June and currently under preview that assists programmers by making relevant suggestions as they write code inside development environments like Microsoft’s Visual Studio Code.
However, citing a recent paper published by OpenAI itself, VentureBeat reports that despite being highly capable the system has “significant limitations”, including biases and sample inefficiencies. According to the paper’s authors, Codex can sometimes propose code that’s syntactically incorrect, and can even call on undefined variables.