Code generation systems like DeepMind’s AlphaCode, Amazon’s CodeWhisperer and OpenAI’s Codex, which powers GitHub’s Copilot service, offer a realistic view of what’s possible with AI in computer programming today. But so far only a handful of such AI systems have been made freely available to the public, reflecting the commercial incentives that open-source companies build them.
To change that, AI startup Embrace Face and ServiceNow Research, ServiceNow’s R&D division, today launched BigCode, a new project aimed at developing “modern” AI systems in “open and responsible” code. road. The goal is to eventually release a large enough data set to train the code generation system, which is then used to create a prototype — a 15-billion parameter model, smaller in size than Codex (12 billion parameters) but larger than Alpha Code (~41.4 billion parameters) — ServiceNow’s home Using a graphics card cluster. In machine learning, metrics are the parts of an AI system that are learned from historical training data and generate code that describes the system’s skills on a problem.
Inspired by Hugging Face BigScience’s effort to open up highly sophisticated text generation systems, BigCode will be open to anyone with a professional AI research background who can devote time to the project, organizers said. The application form went live this afternoon.
“In general, we expect applicants to have a connection to a research organization (in academia or industry) and to work on technical/ethical/legal aspects. [large language models] For code applications,” ServiceNow wrote in a blog post. “Once [code-generating system] It’s trained, we evaluate its capabilities… We try to make the evaluation simple and comprehensive so that we can learn more. [system’s] capacity”
By co-developing an open-source code generation system that allows developers to reuse it under certain terms and conditions, BigCode seeks to address some of the controversies surrounding AI-practice. Powered code generation – especially regarding fair use. The nonprofit Protect Software Freedom and others have criticized GitHub and OpenAI for using public source code, not all of which is licensed, to train and monetize the code. The codex is available through OpenAI’s paid API, and GitHub recently started charging for access to Copilot. For their part, GitHub and OpenAI continue to confirm that Codex and Copilot are not compatible with any license terms.
BigCode’s developers say they take pains to ensure that only files from licensed repositories are included in the specified training dataset. Along the way, he said, he will seek input from relevant stakeholders before issuing policy statements to establish “responsible” AI practices for training and sharing all types of coding systems.
ServiceNow and Hug Face have not provided a timeline for when the project might be completed. But in the next few months, expect it to explore several types of code generation, including systems that automatically compile and compile from code snippets and natural language statements and work across domains, tasks, and programming languages.
Assuming the ethical, technical, and legal issues are one day gone, AI-powered coding tools can dramatically reduce development costs, allowing coders to focus on more creative tasks. According to a study by the University of Cambridge, at least half of the developers’ efforts are spent on debugging and not actively programming, which is estimated to cost the software industry 312 billion dollars a year.
