The role of artificial intelligence-based language models is expanding, and most recently they have also become involved in programming.
SourceAI
French startup SourceAI is using the GPT-3 model to translate English-language requests into computer code in forty programming languages. The company is not alone: others are also trying to make coding easier with artificial intelligence solutions.
Language models are trained to "understand" and guess the needs of programmers.
Image source: https://gptcrush.com.
With SourceAI's solution, which is still in beta test status, users describe the features they want and then choose a programming language. In the current (beta) version, 80-90% of the code generated works as expected. In the future, the company plans to charge between $0.04 USD and $0.1 USD per piece of code.
What is GPT-3?
GPT-3 (the 3 stands for third generation), which generates material that is almost identical to human text and was launched in May 2020, is developed by San Francisco's pioneering AI lab OpenAI. Its text is deceptively similar to human text, and several researchers are alarmed, or even frightened by what they say are too good results.
Debuild
Debuild, powered by GPT-3, creates web applications, such as buttons and text input fields, based on instructions in natural language English.
GPT-2
The Belgian startup Tabnine is working on the previous, second generation of the model (GPT-2) and, while the programmer is typing, it is making automatic suggestions for the next lines.
Large companies that dominate the infocom world and artificial intelligence developments are also using machine learning, language models to increase the creativity of programmers and to automatically filter out errors and bugs.
Facebook? Intel? Google? DeepMind?
With Facebook Aroma, developers can search code databases for code snippets similar to the ones they're working with. Intel's code similarity inference technique is a similar tool - it compares code snippets to determine their functionality. London-based DeepMind, a Google-owned algorithm that beat one of the world's top go-getters a few years ago, rewrites human-developed code to make it more efficient.
Beginner and experienced programmers
In the hands of skilled and talented programmers, these solutions can save a lot of time, leaving more energy for more complex tasks. If they are beginners, and of course if they are attentive enough, they will learn programming faster thanks to the new models.