GH-300 sample exam questions

Copilot Chat combines the model's training data with context from your code and repositories, and can use external sources such as Bing web search for up-to-date answers. It does not rely only on training data, only user documentation, or primarily on Bing results.

Fairness is built in by training on data and algorithms that are checked and corrected for bias, so the system does not treat groups differently. Monitoring (C) helps detect issues but does not by itself ensure fairness; accessibility relates to inclusiveness, and large datasets can still be biased.

The Copilot usage metrics API reports counts such as code suggestions shown and accepted, active users, and Copilot Chat interactions and acceptances, broken down by language and editor. It does not grade code quality, check style compliance, or refactor code.

Suggestions stop when there is no internet connection (Copilot runs in the cloud), when you lack a valid Copilot license or seat, or when content exclusion blocks the file. Files in .gitignore still get suggestions, and Copilot works with essentially all languages, so B is weaker.

A code suggestion's life cycle starts by capturing the user's context (code around the cursor, open files, the prompt), building a prompt, and then generating suggestions with the model before filtering and returning them. Retraining the model, storing user data, and telemetry processing are not part of producing a suggestion.

The GitHub Copilot proxy service, which receives prompts, applies filters and forwards requests to the model, is hosted in Microsoft Azure. It is not hosted on AWS or Google Cloud and cannot be self-hosted.

The most effective way to get better results from Copilot Chat is prompt engineering: be specific, give context and examples, and break problems into smaller questions. A single query for many requirements gives weaker answers, and limiting users or memory use in generated code does not affect Chat quality.

Few-shot prompting gives the model a few examples showing the pattern or mechanism you want, so it applies that approach in its response. It is not asking Copilot to retry or iterate multiple times, nor specifying which sources to use.

In the design phase, Copilot Chat can suggest suitable design patterns, architectures and best practices for the project's requirements. It does not create UI prototypes without prompting, produce a complete design independently, or manage team collaboration and version control.

Copilot reduces boilerplate by suggesting repetitive code, such as constructors, mappings or test scaffolding, based on patterns in your project. It does not refactor the whole codebase without input, collaborate with multiple developers in real time, or predict future requirements.

To follow company testing standards, include specific examples or explicit requirements of your standards in the chat prompt so Copilot mirrors them. A slash command like /tests alone does not know your standards, and no terminal command applies them. Custom instructions files are now another option.

Copilot can only spot missing tests if it sees the relevant code, so the key factor is giving it the right context: open the source and test files, select code, or reference files with #file or @workspace. Closing other tabs removes context, and existing coverage is not the deciding factor.

The /tests slash command asks Copilot Chat to generate unit tests for the selected code. Commands such as /create-tests, /generate-tests or /init-tests do not exist.

To reduce the risk of IP or license infringement, set the suggestions matching public code policy to Block, so Copilot suppresses suggestions that match public code. Allowing public code matches does the opposite, and there is no license check configuration setting.

The duplication detection filter (the suggestions matching public code policy) can be set to Block at the organization level so Copilot does not suggest code matching public code on GitHub. Chat in the IDE or on GitHub Mobile and Bing access are unrelated policies.