Image credits: Rafael Henrique/SOPA Images/LightRocket / Getty Images
Earlier this year, GitHub rolled out Copilot Chat, a chatbot focused on ChatGPT-like programming for organizations subscribing to Copilot for Business. Copilot Chat more recently came to individual Copilot customers – those paying $10 per month – in beta. And now, GitHub is launching Chat generally available for all users.
Starting today, Copilot Chat is available in the sidebar of Microsoft’s IDEs, Visual Studio Code, and Visual Studio – included in paid tiers of GitHub Copilot and free for teachers, students, and verified maintainers of select open projects source.
“As the home of developers around the world, we have brought to market what is now the most widely adopted AI development tool in history,” said Shuyin Zhao, vice president of management products at GitHub, to TechCrunch in an email interview. “And the finished code was just the beginning.”
Not much has changed in Copilot Chat since beta.
The chatbot is still powered by GPT-4, OpenAI’s flagship generative AI model, specially adapted for development scenarios. Developers can ask Copilot Chat in natural language for real-time advice, for example by asking Copilot Chat to explain concepts, detect vulnerabilities, or write unit tests.
Like all generative AI models, the model behind Copilot Chat, GPT-4, was trained on publicly available data, some of which is copyrighted or under a restrictive license. Vendors, including GitHub, say the fair use doctrine protects them from copyright claims. But that hasn’t stopped coders from filing class-action lawsuits against GitHub, Microsoft (GiHub’s parent company), and OpenAI over what they claim are open source licensing and intellectual property violations.
I asked Zhao if code base owners would now have the option to opt out of the training if they wanted. She said there was no new mechanism for this with the wider launch of Copilot Chat and instead suggested that codebase owners make their repositories private to avoid them being included in future training sets.
I have to imagine that codebase owners won’t like this suggestion too much – there are many reasons to keep copyrighted code public, not the least of which is crowdsourced bug hunting. But GitHub clearly isn’t willing to budge on unsubscribing training data – or not yet, at least.
Generative AI models, including GPT-4, also have a tendency to hallucinate or assertively make up facts, which is particularly problematic in the field of coding. According to a recent Stanford study, developers who use AI assistants to code tend to produce less secure code than those who don’t use AI assistants, in part because AI assistants introduce buggy code snippets or obsolete.
Zhao said GPT-4 works “better” against hallucinations compared to the older model that once powered Copilot and pointed to exploit mitigation features such as filters for insecure code patterns, which notify Copilot Chat users experience vulnerabilities such as hardcoded credentials, SQL injections, and path injections. But she stressed the importance of careful human review of any AI-suggested code.
“GitHub Copilot is powered by OpenAI’s models, which we have found to be the best models for the services we offer today,” Zhao said. “We are in a very strong position to continue providing developers with the AI tools they need to build better, more secure software at scale – and to have fun while they do it.”
In October, Microsoft CEO Satya Nadella told analysts that Copilot had 1 million paying users and about 37,000 enterprise customers. But it’s up to GitHub to make Copilot even more attractive, lest it lose ground to competitors — and, for that matter, lose money.
According to a Wall Street Journal article, Copilot loses an average of $20 per month per user, with some customers costing GitHub as much as $80 per month. The high price of running the underlying AI models is believed to be to blame – a problem that coding startup GenAI also encountered, forcing it to shut down early last December.
While GitHub struggles to make Copilot profitable, Amazon continues to upgrade CodeWhisperer, perhaps Copilot’s most well-resourced rival.
In April, Amazon made CodeWhisperer free for developers, with no usage restrictions. That month also saw the launch of CodeWhisperer Professional Tier, which added single sign-on with AWS Identity and Access Management integration as well as higher security vulnerability scanning limits. A business plan for CodeWhisperer was launched in September. And in in early November, Amazon “optimized” CodeWhisperer to provide “Improved” suggestions for developing applications on MongoDB, the open source database management program.
Besides CodeWhisperer, Copilot has competitors in startups like Magic, Tabnine, Codegen, and Laredo, as well as open source models like Code Llama from Meta and StarCoder from Hugging Face and ServiceNow.
Gn En tech