# input (10-input.us)

Input is what someone gives to a process or system. It can be words, numbers, files, or choices entered into a computer, a conversation, or any task.

## What a model may hear
- user prompt text (chat interfaces and LLM APIs): Treat the content as a prompt to be tokenized, processed, and responded to; may apply safety filters or system instructions automatically
- data fed into a function or model (machine learning and programming): Expect structured tensors, arrays, or variables with specific shapes and types; may fail or produce errors if format is wrong
- HTML form field value (web development): Read the value from a DOM element, often as a string, potentially vulnerable to injection if not sanitized
- standard input stream (operating systems and shell scripting): Read bytes from stdin, a pipe, or a redirected file; may block waiting for data if the stream is empty
- economic or resource contribution (economics and business modeling): Incorporate the value as a factor in calculations of production, cost, or output

## Where people and models part ways
- Says: "I gave it bad input."
  Means: I made a mistake in what I typed or said.
  May be taken as: A model may ask for the specific data type, shape, or schema that failed, rather than addressing the content error.
  Say instead: "I made a mistake in what I typed. Let me rephrase."
- Says: "Can you take input from a file?"
  Means: Can you read and use the contents of a file I upload?
  May be taken as: A model may describe how to pipe file contents into stdin or use a file API, rather than using an uploaded document.
  Say instead: "Please read the uploaded file and use its contents."
- Says: "My input is the sales numbers."
  Means: The information I am providing is the sales numbers.
  May be taken as: A model may expect a specific structured data format or variable name called 'input' rather than treating it as conversational context.
  Say instead: "Here are the sales numbers for us to discuss."
- Says: "I need more input on this."
  Means: I want more information, opinions, or discussion.
  May be taken as: A model may interpret this as a request to accept additional data into a function or system, rather than asking for elaboration.
  Say instead: "Can you give me more information or opinions on this?"

## Tips
- Say 'upload' or 'attach' when you mean a file, not 'input'.
- Specify the format you are giving, such as 'Here is a list' or 'Here is a paragraph', to help the model parse correctly.
- If you want a model to process data, describe the data and what you want done, rather than using 'input' as a variable name.
- Avoid saying 'bad input'; say what was wrong in plain language.
- If you mean conversation, say 'my question' or 'my message' instead of 'my input'.

## Often confused with
- prompt: A prompt is specifically the text you send to an LLM; input is broader.
- query: A query is a question or request for data; input is anything fed in.
- feedback: Feedback is response about past output; input is what goes in initially.
- data: Data implies structured, processable facts; input can be any contribution.
- stdin: Stdin is a specific OS stream; input is the general concept.
- argument: An argument is a parameter in code; input is broader and less formal.