input-output prompting

Table of Contents

Fundamental interaction mode with LLM: give a prompt, get an output.

prompt
a text string “who is the president of US in 2014?” feeded into the LLM
output
the response emitted by the LLM “Obama.”

Backlinks

concatenating prompt used in input-output prompting with bunch of examples.

instead of
"what is 3 + 2?"
put
"what is 3 + 2?, 1 + 1 = 2, 2 + 2 = 4, 1 + 2 = 3, 1 + 4 = 5."

If you think “+” would be a common operator and thus part of common human language, replace it with a custom operator like “krl”(“what is 3 krl 2?”).

context here referes to the tokens in the same session as the prompt. in-context learning refers to LLM is learning from the examples we had given in the context(session).

Author: Linfeng He

Created: 2024-04-03 Wed 20:21