The artificial intelligence (AI) research firm OpenAI has unveiled the latest Codex software, transcribing plain English into code for basic games and websites.
The company's CEO, Sam Altman, said via Twitter that the new AI system is a rudimentary version of what would be possible in the future. The company expects the technology to quickly improve over time.
They released a live demo that shows how OpenAI uses the new Codex system to convert written English commands into codes.
Codex Software Could Make Coding Faster
OpenAI has previously shown how Codex software works in previous live demos. According to The Verge, its latest update can translate natural languages into codes that build simple websites and rudimentary games and translate different programming languages and tackle data science queries.
The commands, such as "create a webpage with a menu on the side and title at the top," can be easily translated into codes by Codex. It is a feat that was thought impossible before. However, the technology is nowhere near perfect as it takes some patience to operate. Nonetheless, it proves a valuable tool in making coding faster for web developers.
OpenAI's CTO and co-founder Greg Brockman told The Verge that this new tool can make the second step of programming faster and remove its most tedious part.
Before Codex, OpenAI used the tool called Copilot, which is an earlier version of Codex. It offers suggestions on how to finish lines but only in codes as users type them out. However, the new Codex is much more advance and flexible as it can create codes and not just complete them.
Codex is built on OpenAI's GPT-3 language generation model trained on a sizable chunk of the internet. One application for GPT-3 was to generate codes, but Codex also exceeded its predecessor because it is specifically trained on open-source code repositories that came from the web.
Limitations and Biases of Codex
Codex is an impressive AI system of OpenAI, but that does not mean it is perfect. According to The Next Web's report, the upgrade of Codex also has its limitations and biases.
"The code neural networks that you will have in the future will be far better than this," said OpenAI Chief Scientist Ilya Sutskever, as quoted by TNW. "This is only the beginning of an exciting future."
In their recently published paper, titled "Evaluating Large Language Models Trained on Code," they highlighted the things that are out of the AI system's scope. Nevertheless, Codex's simplicity and adaptability provide a promising foundation for artificial intelligence coding assistants.
Some programmers see Codex as a potential threat to their jobs, while others envision it as taking erroneous tasks out of their hands so they could focus on developing new ideas.
RELATED ARTICLE: OpenAI Deceived? Handwritten Notes Fool Computer Vision in Typographic Attack