Instruction: A specific task/instruction you want model to perform
Context: Additional information
Input Data: Input/Question we want response for
Output Indicator: the type or format of the output
Not required in instruction tuned models
Example
Classify the text into neutral, negative, or positive // Instruction
Text: I think the food was okay. // Input Data
Sentiment: // Output Indicator
Prompting Techniques
Zero Shot Prompting
It means that the prompt won’t contain examples or demonstrations
Instruction Tuning has improved zero shot prompting
When Zero Shot does not work, use Few Shot Prompting
Few Shot Prompting
It means that the prompt will contain some examples or demonstrations
1-shot: 1 example
2-shot: 2 examples
Limitation
Not good with complex reasoning tasks
A "whatpu" is a small, furry animal native to Tanzania. An example of a sentence that uses the word whatpu is:
We were traveling in Africa and we saw these very cute whatpus.
To do a "farduddle" means to jump up and down really fast. An example of a sentence that uses the word farduddle is:
Chain-Of-Thought Prompting
With Zero Shot
Just add Let's think step by step at the end of prompt
I went to the market and bought 10 apples. I gave 2 apples to the neighbor and 2 to the repairman. I then went and bought 5 more apples and ate 1. How many apples did I remain with?
Let's think step by step.
With Few Shot
Give reasoning steps in the examples
we can get better results on more complex tasks that require reasoning before responding
The odd numbers in this group add up to an even number: 4, 8, 9, 15, 12, 2, 1.
A: Adding all the odd numbers (9, 15, 1) gives 25. The answer is False.
The odd numbers in this group add up to an even number: 15, 32, 5, 13, 82, 7, 1.
A:
ReAct
Reasoning+Acting
It allows the system to perform dynamic reasoning to create, maintain, and adjust plans for acting while also enabling interaction to external environments
Langchain latest implementation do not use ReAct loop
Thought 1: I need to search Apple Remote and find the program it was originally designed to interact with
Act 1: Search[Apple Remote] // Tool call
Observation 1: <>
Thought 2: <>
Act 2: Search[Front Row]
Observation 2: <>
Thought 3: <>
Act 3: Finish[Keyboard Function Keys] // Final answer