Earlier this year, GitHub launched Copilot Chat, a programming-centric chatbot similar to ChatGPT for organizations that subscribe to Copilot for Business. Copilot Chat recently launched in beta for individual Copilot customers (paying $10 per month). Now, GitHub is fully rolling out Chat to all users.
As of today, Copilot Chat is available in the Microsoft IDE, Visual Studio Code, and the sidebar of Visual Studio – as part of the GitHub Copilot premium plan, and free for verified teachers, students, and maintainers of certain open source projects.
“As the home of developers around the world, we have brought to market the most widely adopted artificial intelligence developer tools in history,” Shuyin Zhao, vice president of product management at GitHub, told TechCrunch in an email interview. “Code completion is just the beginning.”
The rest of the co-pilot chat has changed little since the beta.
The chatbot is still powered by OpenAI’s flagship generative artificial intelligence model GPT-4, which is fine-tuned specifically for development scenarios. Developers can prompt Copilot Chat with natural language to get instant guidance, such as asking Copilot Chat to explain a concept, detect a vulnerability, or write a unit test.
Like all generative AI models, the model that powers Copilot Chat, GPT-4, was trained on publicly available data—some of which is copyrighted or subject to restricted licenses. Vendors, including GitHub, argue that fair use protects them from copyright claims. But that hasn’t stopped programmers from filing class-action lawsuits against GitHub, Microsoft (GiHub’s parent company) and OpenAI, alleging open source licensing and intellectual property infringement.
I asked Zhao if there was now an opportunity for codebase owners to opt out of training if they wanted to. She said there are no new mechanisms to address this issue as Copilot Chat becomes more widely available, but instead recommends codebase owners make their repositories private to prevent them from being included in future training sets.
I have to imagine that code base owners won’t take this advice too kindly – there are many reasons to keep copyrighted code public, not the least of which is crowdsourcing bug hunting. But GitHub is apparently unwilling to budge on the training material opt-out—at least not yet.
Generative AI models, including GPT-4, also have a tendency to hallucinate or confidently make up facts, which is particularly problematic in the coding world. Developers who code using AI assistants tend to produce less secure code than developers who don’t use AI assistants, in part because of bugs introduced by AI assistants, according to a recent study from Stanford University or deprecated code snippets.
Zhao said GPT-4 performed “better” against hallucinations than the older models that once powered Copilot, pointing to exploit-mitigation features, such as filters for insecure code patterns, that Features can notify Copilot Chat users of vulnerabilities such as hard-coded credentials, SQL injection, and exploits. Path injection. But she stressed the importance of close human review of any code suggested by AI.
“GitHub Copilot is powered by OpenAI’s model, which we found to be the best model for the services we provide today,” Zhao said. “We are in a great position to continue to provide developers with the AI tools they need to build better, more secure software at scale, and to have fun while doing it.”
In October, Microsoft Chief Executive Satya Nadella told analysts that Copilot had 1 million paying users and about 37,000 enterprise customers. But GitHub has a responsibility to make Copilot more attractive so as not to lose cash to its competitors.
According to the Wall Street Journal, Copilot is losing an average of $20 per user per month, and some customers are losing as much as $80 per month to GitHub. The high cost of running the underlying AI models was reportedly the culprit — an issue also encountered by GenAI coding startup Kite, which forced it to shut down in early December.
While GitHub works to make Copilot profitable, Amazon continues to upgrade CodeWhisperer, perhaps Copilot’s most well-resourced competitor.
In April, Amazon made CodeWhisperer available to developers for free with no usage restrictions. Also launched that month was the CodeWhisperer Professional Tier, which added single sign-on integration with AWS Identity and Access Management and higher limits on security vulnerability scanning. CodeWhisperer’s enterprise program launches in September.and in In early November, Amazon “optimized” CodeWhisperer supply Recommended “enhancements” for application development on the open source database manager MongoDB.
In addition to CodeWhisperer, Copilot’s competitors include startups such as Magic, Tabnine, Codegen and Laredo, as well as open source models such as Meta’s Code Llama and Hugging Face and ServiceNow’s StarCoder.
