GitHub Copilot Chat Fully Launched to Assist Developers with Programming Challenges

Last year, GitHub launched Copilot Chat, a programming-oriented chatbot similar to ChatGPT, specifically designed for organizations subscribed to Copilot for Business. Copilot Chat recently released a beta version for individual Copilot users, who pay $10 per month. Now, GitHub announces the general availability of Chat for all users.

Currently, Copilot Chat is available in Microsoft's IDE, Visual Studio Code, and the sidebar of Visual Studio, as part of the GitHub Copilot paid tier, and is free for verified educators, students, and maintainers of certain open-source projects.

Zhao Shuyin, Vice President of Product Management at GitHub, stated in an email to TechCrunch, "As the home for global developers, we have launched the most widely adopted AI developer tool on the market today." She said, "Code completion is just the beginning."

Since the beta version, Copilot Chat has not undergone significant changes.

This chatbot is still powered by GPT-4, OpenAI's flagship generative AI model, specifically fine-tuned for development scenarios. Developers can use natural language prompts to interact with Copilot Chat and receive real-time guidance, such as requesting explanations of concepts, detecting vulnerabilities, or writing unit tests.

Like all generative AI models, the model GPT-4 supporting Copilot Chat is trained on publicly available data, some of which is protected by copyright or restrictive licenses. Suppliers, including GitHub, believe that fair use principles protect them from copyright claims. However, this has not stopped coders from filing a collective lawsuit against GitHub, Microsoft (GitHub's parent company), and OpenAI, alleging open-source license and intellectual property infringement.

When asked if code repository owners have the opportunity to opt out of training now, Ms. Zhao stated that with the wider release of Copilot Chat, there have been no new mechanisms added. Instead, she suggested that code repository owners make their repositories private to prevent them from being included in future training sets.

Clearly, GitHub is currently unwilling to compromise on the issue of opting out of training data selection.

Generative AI models, including GPT-4, also have a tendency for illusions. According to a recent study by Stanford University, developers who code with AI assistants produce less secure code compared to those who do not use AI assistants, partly due to the introduction of erroneous or outdated code snippets by the AI assistants.

Ms. Zhao stated that GPT-4 performs "better" in combating illusions compared to the old models that powered Copilot, and highlighted some features that mitigate vulnerabilities, such as filters for insecure code patterns. These filters notify Copilot Chat users of vulnerabilities, such as hard-coded credentials, SQL injection, and path injection. However, she emphasized the importance of rigorous human review of any AI-suggested code.

"GitHub Copilot is powered by OpenAI's models, and we find these models to be the best models for the services we currently provide," said Ms. Zhao. "We are in a very favorable position to continue empowering developers with AI tools to help them build better and safer software at scale, while enjoying the process."

In October, Microsoft CEO Satya Nadella told analysts that Copilot has 1 million paid users and approximately 37,000 enterprise customers. However, GitHub must make Copilot more appealing to avoid losing its position in the competition.

According to a report by The Wall Street Journal, Copilot incurs an average monthly loss of $20 per user, with some customers causing losses as high as $80 per month. The high cost of running the underlying AI models is rumored to be the main culprit, which is also a problem faced by GenAI coding startup Kite, forcing it to shut down in early December last year.

While GitHub strives to make Copilot profitable, Amazon continues to upgrade its most resourceful competitor, CodeWhisperer.

In April, Amazon offered CodeWhisperer for free to developers with no usage restrictions. In the same month, they also launched CodeWhisperer Professional Tier, which added single sign-on integration with AWS Identity and Access Management and higher limits for security vulnerability scanning. In September, they introduced the enterprise plan for CodeWhisperer. In early November, Amazon "optimized" CodeWhisperer and provided "enhanced" suggestions for MongoDB, an open-source database management program.

In addition to CodeWhisperer, Copilot also faces competition from startups such as Magic, Tabnine, Codegen, and Laredo, as well as open-source models like Meta's Code Llama and Hugging Face's StarCoder with ServiceNow.