The Chat feature within developer tool GitHub Copilot sometimes suffers from a persistent problem: the user. Overly vague questions prevent the AI-powered chatbot from being able to help them, but a recent update addresses this.
GitHub Copilot Chat will now let users know if their question is ambiguous or otherwise vaguely worded. If the questioner writes down “what is this,” the Chat feature will now tell how it can be worded differently. This will then be the context Copilot will use to arrive at a targeted answer.
Problem solved
Although vague prompts come from the end user, this evolution is also useful for GenAI solutions. After all, LLMs have an annoying tendency to always provide an answer, even if they don’t actually possess the information needed. Detecting such hallucinations is a major pain point and is regularly seen as a fundamental flaw of the technology. Nevertheless, detecting erroneous or questionable prompts has already prevented many issues, usually through AI guardrails. Now, the solution is switched around to improve prompting.
Tip: GitHub Copilot Chat makes AI programming assistance even more nimble
Microsoft is making these changes based on feedback. Customers have realized the usefulness of the Chat feature, but mostly lack proportionality. “I wish it felt more like an assistant,” mentions a GitHub Copilot user quoted by Microsoft. “It sometimes feels like I have to sift through a lot to get to what I actually wanted,” says another.
Productivity boost
The promise of AI-driven programming assistance is to provide a productivity boost. Microsoft is convinced that this improvement actually makes programmers more efficient, more engaged and stop them from being overwhelmed. To keep improving in this area, however, Microsoft is asking for ongoing feedback. GitHub Copilot, which has been around since late 2021, has already received a major boost in the form of OpenAI’s GPT models since the advent of ChatGPT.
In addition to the improved Chat feature, GitHub also recently added support for more LLMs. Thus, the developer tool now also runs Anthropic’s Claude 3.5 Sonnet, Google’s Gemini 1.5 Pro and OpenAI’s o1 preview.
Also read: GitHub Spark builds web apps without code