Codex

Code is the language that every computer uses to ‘speak’ and ‘understand.’ When developers write code, they make it to issue instructions to a computer and tell them what to do. But as technology has evolved and everything including the way we order food to eject satellites into space has moved to digital mode, the need for coding has drastically surged. But despite its increasing demand, we are still not able to take coding to laymen. When it comes to programming languages and coding, only tech-savvy can get their hands on it. But maybe not anymore! OpenAI has introduced Codex, an AI system that can convert the natural language to programming code.

Due to the prevalence of business applications, learning to write code has become one of the most-sought-after skills in the job market. But not everybody can be well-versed in coding. We can’t demand a telecom company CEO to code. However, he/she might have innovative application ideas that need coding to be streamlined. Therefore, the digital world is making coding easy by introducing no-code or low-code applications. As a further move, OpenAI has taken an initiative to convert the natural language to programming code. In this article, IndustryWired takes you through how Codex is a breakthrough innovation that could change the fate of programming.

Codex, an Innovative Move to Streamline Coding

Codex, the free API model released by OpenAI, is not a new invention. The company has been working on innovative initiatives related to coding for quite some time now and Codex is one of the latest developments that is aimed at taking natural language to programming code. Built on Copilot, a tool for use with Microsoft’s GitHub code repository, Codex gives suggestions for coding just like how Google ends all our sentences. Similarly, Copilot finishes the lines of code.

By adding this existing feature to Code, it makes the AI system even more sophisticated. Taking the initiative even further, Codex accepts sentences written in English and translates them to runnable code. For example, when a user asks Codex to formulate a new application according to his/her needs, the AI system grabs the concept and converts the natural language commands to create the actual application. Besides, the system can both generate and parse code, which allows users to get a customized coding end. 

The Downside of Codex

A recent paper published by OpenAI has stated that although Codex is an extraordinary AI system that could bring programming to the hands of non-tech people, it has downsides like bias and sample inefficiencies that need to be addressed. While its suggestion option is often applauded, on the other hand, the model proposals are found to be syntactically incorrect or undefined code, invoking variables and attributes that are undefined or outside the scope of the codebase. 

Like many other applications, Codex also develops suggestions based on the data it is trained with. Therefore, we are back to square one where we talked about AI models being trained with biased data. It leads to racist and harmful outputs as code. For example, Codex uses harmful words like ‘terrorist’ and ‘violent’ while writing codes related to ‘Islam.’ 

Besides, OpenAI also drew criticism for using programming code written by other researchers for the GitHub repository. Many of those who contributed to GitHub accused OpenAI of using their code for profit.