AI terms in plain words

A

• Agentic workflow: An agentic workflow is a process in which an AI model plans steps, uses tools and checks its own progress to finish a task.
• AI agent: An AI agent is a program that uses an AI model to plan steps and act on its own toward a goal you set, such as sorting messages or filling a form.
• AI assistant: An AI assistant is a chat-style program that answers questions and writes or edits text, images or code when you ask it.
• AI avatar: An AI avatar is a computer-generated presenter that speaks a script on video, usually with an AI voice.
• AI labelling (AI disclosure): AI labelling is telling people clearly when a voice, picture, video or text was made with AI.
• AI literacy: AI literacy is knowing enough about how AI works, what it does well and where it fails to use it safely and sensibly.
• API: An API is a way for one program to talk to another so they can exchange data or trigger actions without a person clicking.
• API key: An API key is a secret code that identifies you to a service and lets your programs use it on your account.

• Benchmark: A benchmark is a standard test used to compare AI models on the same set of tasks.
• Bias in AI: Bias in AI is a pattern in a model's answers that unfairly favours or disfavours some people or views, usually inherited from its data.

• Chain of thought: Chain of thought is a way of getting an AI model to work through a problem step by step before it gives the final answer.
• Chatbot: A chatbot is a program that holds a conversation with people, answering questions in text or voice.
• Citation: A citation is a pointer from a claim to the source it came from, such as a link, a page or a document name.
• Closed model: A closed model is an AI model that you can use only through its maker's app or service, without access to its files.
• Commercial use: Commercial use means using what a tool produced to make money or support a business, which some plans allow and others restrict.
• Content credentials: Content credentials are tamper-evident information attached to a file that records who made it and how, including whether AI was used.
• Context window: A context window is the amount of text an AI model can take in at once, including your instructions, your documents and its own earlier answers.
• Copyright and AI: Copyright and AI is the set of open questions about who owns what an AI tool creates and whether the material it learned from was used properly.
• Cost per use: Cost per use is the price of one unit of work with an AI tool, such as one generated clip, one minute of audio or one batch of text.
• Credits: Credits are units a tool uses to measure how much you can generate, with different actions costing different amounts.

• Data privacy: Data privacy is protecting personal and confidential information so that only the people and systems that should see it can.
• Data retention: Data retention is how long a service keeps the data you give it, such as prompts, uploads and results.
• Deepfake: A deepfake is a realistic fake image, video or audio of a real person made with AI to show something they did not do or say.

• Embeddings: Embeddings are lists of numbers that represent the meaning of a piece of text, an image or a sound so that similar things end up close together.
• Evaluation: Evaluation is testing how well an AI system does a task by checking its answers against a set of cases with known good results.

• Few-shot prompting: Few-shot prompting is giving an AI model a small number of worked examples inside your prompt so it copies their pattern and style.
• Fine-tuning: Fine-tuning is training an existing AI model further on your own examples so that it answers in a style or domain you need.
• Free plan: A free plan is a version of a tool you can use without paying, usually with limits on credits, length, quality or features.
• Free trial: A free trial is a short period of full or partial access to a paid plan before you decide to pay.

• GDPR basics: The GDPR is a European law that sets rules for how organisations collect, use, store and delete personal data of people in the European Union.
• Generative AI: Generative AI is AI that creates new content, such as text, images, audio or video, from a request instead of only analysing existing data.
• Grounding: Grounding is tying an AI model's answer to specific sources you provide, so it uses them instead of guessing from memory.
• Guardrails: Guardrails are rules and checks placed around an AI system to keep its answers and actions inside safe, intended limits.

• Hallucination: A hallucination is an answer from an AI model that sounds confident but is false, invented or unsupported by any source.
• Human in the loop: Human in the loop means a person reviews or approves an AI system's output or action before it takes effect.

• Image prompt: An image prompt is a written description, sometimes with a reference picture, that tells an AI tool what image to create.
• Image-to-video: Image-to-video is AI that adds motion to a still picture to make a short clip.
• Inference: Inference is the moment an already trained AI model is used to produce an answer, as opposed to training it.

• Jailbreak: A jailbreak is a prompt written to trick an AI model into breaking the safety rules its makers set.

• Large language model (LLM): A large language model is an AI model trained on a very large amount of text to predict and produce language, which is how chat assistants write answers.
• Latency: Latency is the delay between sending a request to an AI service and getting the answer back.
• Lip sync: Lip sync is matching the mouth movements of a person or character in a video to a spoken audio track.
• Low-code: Low-code is building software mostly with visual blocks while allowing small pieces of written code where blocks are not enough.

• Model Context Protocol (MCP): The Model Context Protocol is an open standard that lets an AI assistant connect to outside tools and data through a common interface.
• Multimodal AI: Multimodal AI is AI that can work with more than one kind of input or output, such as text, images, audio and video.
• Multimodal prompt: A multimodal prompt is an instruction that combines more than one kind of input, such as text together with a photo, a document or a recording.

• Negative prompt: A negative prompt is a list of things you tell an image or video tool to leave out of the result.
• No-code: No-code means building apps, forms or automations by clicking and arranging blocks instead of writing programming code.

• Open-source model: An open-source model is an AI model whose files are shared so that others can download, run and often adapt it under a stated licence.
• Orchestration: Orchestration is coordinating several AI models, tools and steps so they run in the right order and pass results to each other.
• Overfitting: Overfitting is when an AI model learns its examples too closely and then does badly on new cases it has not seen.

• Prompt: A prompt is the instruction you give an AI model, including the task, the facts it should use and the format you want back.
• Prompt injection: Prompt injection is an attack where hidden or tricky text makes an AI model ignore its instructions and do something else.
• Prompt library: A prompt library is an organised collection of tested prompts that you or your team can reuse.
• Prompt template: A prompt template is a reusable prompt with blanks, such as [audience] or [facts], that you fill in each time.

• Rate limit: A rate limit is a cap on how many requests you can send to a service in a given period.
• Red teaming: Red teaming is deliberately trying to break or mislead an AI system to find its weak points before real users do.
• Retrieval-augmented generation (RAG): Retrieval-augmented generation is a method where the AI first looks up relevant passages from your documents and then writes its answer using them.

• Seed: A seed is a number that starts the random choices of an image or video generator, so the same seed with the same settings can give a similar result again.
• Subscription vs credits: Subscription versus credits is the choice between paying a regular fee for a plan and paying for a bundle of credits that you spend as you generate.
• Summarisation: Summarisation is shortening a long text or recording into its main points, which AI tools can do on request.
• Synthetic media: Synthetic media is any image, video, audio or text that was generated or heavily altered by AI instead of being recorded or written directly.
• System prompt: A system prompt is a standing instruction given to an AI model before the conversation starts, which sets its role, tone and rules.

• Temperature: Temperature is a setting that controls how predictable or varied an AI model's wording is, with low values giving steadier answers and high values giving more surprising ones.
• Terms of service: Terms of service are the rules a company sets for using its product, including what you may do, who owns the output and how your data is treated.
• Text-to-speech (TTS): Text-to-speech is technology that turns written text into spoken audio with a synthetic voice.
• Text-to-video: Text-to-video is AI that creates a short video clip from a written description or a script.
• Token: A token is a small piece of text, often part of a word, that an AI model reads and writes, and the unit many tools use to measure limits and cost.
• Tokenization: Tokenization is the step where text is cut into tokens, the small pieces an AI model actually reads.
• Tool use (function calling): Tool use, also called function calling, is when an AI model asks another program to do something for it, such as search, calculate or send data, and then uses the result.
• Training data: Training data is the collection of text, images, audio or other examples an AI model learned from.
• Transcription: Transcription is turning spoken audio or video into written text, which AI tools can now do automatically.

• Upscaling: Upscaling is using software to enlarge a picture or video and add plausible detail so it looks sharper at a bigger size.

• Vector database: A vector database is a store built to save embeddings and quickly find the items whose meaning is closest to a question.
• Voice cloning: Voice cloning is creating a synthetic voice that sounds like a specific real person from recordings of that person.

• Watermark: A watermark is a visible mark or logo added to a picture or video, often on free plans of AI tools.
• Watermark detection: Watermark detection is using software to find an invisible or visible signal that shows a picture, video or audio was made by an AI tool.
• Webhook: A webhook is an automatic message one program sends to another when something happens, such as a new form answer.
• Workflow automation: Workflow automation is setting up steps to run on their own when something happens, such as a new form answer or a new recording.

• Zero-shot prompting: Zero-shot prompting is asking an AI model to do a task with instructions only, without showing it any examples first.

GuidesTools