Copilot Chat is a programming-centric chatbot that was released by GitHub earlier this year. It is similar to ChatGPT and is designed for enterprises that have subscribed to the Copilot for Business service. More recently, the beta version of Copilot Chat was made available to individual Copilot users who pay a monthly fee of ten dollars. Additionally, GitHub is now making Chat available to all users in a general capacity.
Copilot Chat is now accessible in the sidebar of Microsoft’s integrated development environments (IDEs), Visual Studio Code and Visual Studio. It is also included as a component of the commercial tiers of GitHub Copilot, and it is free for verified teachers, students, and maintainers of certain open source projects.
“As the home of the world’s developers, we’ve brought to market what is now the most widely adopted artificial intelligence developer tool in history,” said Shuyin Zhao, vice president of product management at GitHub, in an email interview with Xyztech. Moreover, the completion of the code was only the beginning.
There have been very few other changes made to Copilot Chat since the beta version.
OpenAI’s flagship generative AI model, GPT-4, which has been fine-tuned specifically for development settings, continues to be the source of power for the chatbot. It is possible for developers to ask Copilot Chat questions using natural language in order to receive real-time help. For instance, developers can ask Copilot Chat to clarify concepts, identify vulnerabilities, or construct unit tests.
Like all other generative AI models, the model that underpins Copilot Chat, known as GPT-4, was trained on data that is freely accessible to the public. However, some of this data is protected by copyright or is licensed under a restrictive license. The fair use theory, according to some vendors, protects them against copyright disputes. GitHub is one of these vendors. However, this has not prevented coders from initiating class action lawsuits against GitHub, Microsoft (the parent company of GiHub), and OpenAI, alleging that these companies have violated their intellectual property rights and committed open source licensing violations.
I inquired with Zhao about whether or not owners of codebases will be given the opportunity to opt out of training at this time, in the event that they desire to do so. As a result of the larger rollout of Copilot Chat, she stated that there is no new mechanism for this, and instead, she proposed that owners of codebases make their repositories private in order to prevent them from being included in future training sets.
I have to imagine that owners of codebases will not take that notion very well. There are numerous reasons for keeping code that is protected by copyright public, and crowdsourcing bug hunting is not one of them. GitHub, on the other hand, does not appear to be willing to move on the issue of opting out of training data, at least not yet.
Additionally, generative artificial intelligence models, such as GPT-4, have a propensity to hallucinate, which means they confidently make up facts. This is a particularly problematic condition in the field of coding. A recent study conducted at Stanford found that developers who use artificial intelligence assistants to write code tend to generate code that is less secure than code written by developers who do not use AI assistants. This is due, in part, to the fact that AI assistants present code snippets that are either deprecated or include bugs.
According to Zhao, GPT-4 shows “better” performance against hallucinations when compared to the older model that was used to power Copilot. He also mentioned exploit-mitigating features such as filters for insecure code patterns, which alert users of Copilot Chat to vulnerabilities such as hardcoded credentials, SQL injections, and path injections. On the other hand, she emphasized the significance of human inspection scrutinizing any code that was suggested by AI.
Zhao stated that OpenAI’s models, which we have determined to be the most suitable models for the services that we provide at the present time, are the ones that power GitHub Copilot. “We are in a really strong position to continue empowering developers with the artificial intelligence tools they need to build better, more secure software at scale, and to have fun while they are doing it,” said the company.
Copilot had approximately 37,000 enterprise clients and one million paying users, according to statements made by Satya Nadella, the CEO of Microsoft, to analysts in October. On the other hand, it is GitHub’s responsibility to make Copilot even more appealing in order to prevent the company from falling behind its rivals and, for that matter, losing money.
There are certain users who cost GitHub as much as $80 each month, according to an article that was published in the Wall Street Journal. Copilot loses an average of $20 per user by the end of each month. According to reports, the high cost of running the underlying AI models is to blame. This is exactly a problem that the GenAI coding startup Kite had, which ultimately led to the company’s closure at the beginning of December of last year.
While GitHub is having trouble turning Copilot into a lucrative business, Amazon is continuing to improve CodeWhisperer, which is possibly the most resource-dense competitor to Copilot.
Over the course of the month of April, Amazon made CodeWhisperer available to developers at no cost and without any usage restrictions. This month also saw the launch of CodeWhisperer Professional Tier, which included higher restrictions on scanning for security vulnerabilities and single sign-on interaction with Amazon Web Services Identity and Access Management. All of these features were added during that month. In September, an enterprise plan for CodeWhisperer was introduced to the public. And in the beginning of November, Amazon “optimized” CodeWhisperer so that it could offer “enhanced” recommendations for the construction of applications using MongoDB, which is an open-source database management program.
Additionally, in addition to CodeWhisperer, Copilot has competition from startups such as Magic, Tabnine, Codegen, and Laredo, as well as open source models such as Meta’s Code Llama and Hugging Face’s and ServiceNow’s StarCoder.