Zero-shot Multi-Shot prompting
0-Shot, 1-Shot, 2-Shot, 3-Shot, 4-Shot, 5-Shot are all a way of prompt engineering, using examples to shape the response from an LLM.
"Shot" terminology is something you will see in model comparisons for benchmarking and can be confusing when trying to chose models to use for an application use case. In fact it is pretty straight forward and an important thing to learn in order to create the best prompts when engineering prompts.
0-shot prompting
0-shot learning
These two terms can be used interchangably. The term "learning" is about teaching the LLM within the prompt by giving it examples of output. By doing so it is learning within the prompt. This duality of terminology naming threw me intially until I realised its the same thing.
0-Shot, 1-Shot, 2-Shot, 3-Shot, 4-Shot, 5-Shot are all a way of prompt engineering, using examples to shape the response from an LLM
Prompting
When prompting an LLM, it can be helpful to give the model an expample of the type of output required, making it more likely to output the desired output. This technique of giving examples is called Zero or Multi-Shot prompting.
Here are some examples:
0-shot prompt
No example is given, you entrust the LLM to return a suitable response from its knowledge.
Note the extra output and comprehensive answer the LLM provides.
1-shot prompt
A single example is provided of the output desired for the prompt.
Note how it is now horizontal and a simple response vs zero shot, due to following the example it was given.
Multi-shot prompt
Give the model multiple examples of the form the response should be in. Giving a range of response types can help with complex responses to better tailor the response to desired outcome. A single example may not cover the bredth of possible responses otherwise.
Multiple examples given in the contrived example, but you get the idea, a more complex output is sought.
Can by style or other diections by example too
But it doesn't have to just be the format, it can be directing the content too, say giving examples of the style of the writing given for different outputs.
Model benchmarking
So we now understand these performance benchmarks will have been done with different prompt examples.
This basic overview I hope helped introduce some of the concepts and terminology that you will encounter when getting started with AI. Is is part of a series of posts on AI concepts.