SourceAI

SourceAI Official Website, Text AI Programming Tool, Simplify, Detect and Fix Errors, and Debug Code

What is SourceAI?

SourceAI is an AI-powered tool that can generate code in any programming language from any human language description. It can also simplify, detect and fix errors, and debug code.

SourceAI Official Website:https://sourceai.dev/

History and Development

Writing good code requires years of learning. However, a startup company in Paris called SourceAI believes that programming should not be such a big deal.

The company is fine-tuning a tool using AI technology that can write code based on a brief description of a coding task. For example, if you tell this tool "multiply two numbers given by the user," it will write a few lines of Python code to complete the task.

SourceAI's vision is a sign of a broader revolution in software development. Advanced machine learning can now automatically perform a set of programming tasks, from auto-completing code snippets to fine-tuning algorithms that search for source code and locate pesky bugs.

Automated programming can revolutionize software development, but the limitations and blind spots of modern AI may also bring new problems. Machine learning algorithms may be unpredictable, and machine-generated code may introduce harmful bugs unless carefully checked.

SourceAI and another similar program aim to capitalize on the momentum of GPT-3, a powerful AI language program released by OpenAI in May 2020, a company based in San Francisco dedicated to fundamental developments in AI. The creators of SourceAI were among the first few hundred people to use GPT-3. OpenAI did not release the code for GPT-3 but allowed some users to access the model through an API.

GPT-3 is a massive artificial neural network trained on a vast amount of text scraped from the internet. It does not understand the meaning behind the text, but it can capture language patterns well, generating articles on a given topic, summarizing an article, or answering questions about the content of a document.

"We realized during our testing of this tool that it can generate code," said Furkan Bektes, the founder and CEO of SourceAI, "that's when we had the idea to develop SourceAI."

He was not the first person to notice this potential. Soon after the release of GPT-3, a programmer demonstrated that it could create custom web applications, including buttons, text input boxes, and colors, by recombining input code snippets. Another company, Debuild, is preparing to commercialize this technology.

SourceAI aims to enable users to generate programs written in different languages for a wider range of applications, helping us to automatically create more software. "Developers save time on coding, and people without programming knowledge can also develop applications," Bektes said.

Another company, TabNine, used the previous version of the OpenAI language model, GPT-2, which was also released by OpenAI. It can be used to build tools that provide auto-completion of code lines or functions as developers type.

Some software giants seem to have taken an interest. Microsoft invested $1 billion in OpenAI in 2019 and agreed to a license for GPT-3. At the software giant's build conference in May, OpenAI co-founder Sam Altman demonstrated the process of GPT-3 automatically generating code for developers.

Brendan Dolan-Gavitt, an assistant professor in the Department of Computer Science and Engineering at New York University, said that language models like GPT-3 are likely to be used to help human programmers. Another product that will use this model is "to point out bugs in your code by finding content that surprises language models."

However, using AI to generate and analyze code may pose problems. In an online paper published in March, researchers at MIT revealed that AI programs used to verify the safety of code can be deceived by making careful changes, such as replacing certain variables, to create harmful programs. Ph.D. student Shashank Srikant believes that AI models should not be over-relied upon, "Once these models enter production, the situation could quickly turn dire," he said.

Professor Dolan-Gavitt of New York University said that the language models used in code generation tools inherently have issues. "I believe that using language models directly is likely to result in insecure code riddled with bugs, as they are trained on human-written code, which is often full of bugs and not secure."