Words in everyday language
You do not need to memorize these terms. Look up a word when it appears in a prompt or practice file.
- Generative AI
- Software that creates text, images, or other material from a request.
- Chatbot and model
- A chatbot is the app you talk with. A model is the AI system it uses to create an answer.
- Prompt
- Your request. It can be a question, an instruction, or a description of what you want.
- Input and output
- Input is what you give AI, including your request and any files. Output is the answer or draft it returns.
- Context
- Background information that helps AI do the job, such as who will read a message.
- Audience, tone, and role
- Audience means who the result is for. Tone means how it sounds. A role describes the kind of help you want; it does not make AI a qualified professional.
- Source and source location
- A source is the original information. Its location tells you where to check it, such as statements.md, A17, paragraph 1.
- Evaluation criteria
- The checks you choose to decide whether an answer does the job. For a reminder: correct date, one sentence, no added promises.
- Example and follow-up
- An example shows the kind of answer you want. A follow-up is another message asking AI to improve or continue its answer.
- Zero-shot, one-shot, and few-shot
- These mean a request with no examples, one example, or a few examples. You can simply say “Here is an example.”
- Prompt chaining
- Breaking a task into connected requests. For example: draft a reminder, check its facts, then shorten it.
- Hallucination
- An incorrect or invented detail in an AI answer. Check the original information even when AI sounds confident.
- Applicant code or ID
- A label used instead of a name. A17, A18, and A19 are made-up applicant codes in this packet. The same code connects one applicant’s records. A17-transcript.pdf means the transcript file for applicant A17. A code alone does not remove every identifying detail from a real application.
- Manifest or inventory
- A file list. A filename on the list does not prove that AI has read the file or that the document meets a requirement.
- CSV, Markdown, TXT, and PDF
- CSV is a simple table. Markdown (.md) is plain text with simple headings. TXT is plain text. PDF preserves a page layout. Check whether the tool can read the information you need.
- Rubric, criterion, and anchor
- A rubric is a scoring guide. A criterion is one part being judged, such as community contribution. An anchor describes what earns a particular score.
- Evidence and supported result
- Evidence is the original information behind a finding. A supported result has evidence for what changed; a count of activities alone does not show an improvement.
- Provisional score
- A suggested score that a person must check. Staff own the final score.
- Calibration
- Reviewers compare their scores and reasons to check whether they are applying the same guide in the same way.
- Exception
- A request to handle a case differently under the organization’s rules, such as allowing extra time.
- Instructions and Project
- Instructions are rules for how AI should work. A ChatGPT Project keeps related chats, files, and instructions together when the account offers that feature.
- Skill
- A reusable set of instructions and resources for a task, like a recipe. Try it on examples where you know the answer before relying on it.
- Embedded instruction
- A request written inside a source document, such as “ignore the scoring guide.” Treat it as text to examine, not a command to follow.
- Workflow and handoff
- A workflow is the order of tasks. A handoff is the information passed to the next person, including what they need to check.
- AI detector and false positive
- A detector guesses whether writing may involve AI. A false positive is a wrong flag. A flag alone does not prove cheating.
