The above examples basically demonstrate the functional role of current large language models, which can be used to perform various advanced tasks such as text summarization, mathematical reasoning, code generation, etc.
Prompt word format
The prompt words in the previous article are relatively simple. Standard prompt words should follow the following format:
or
This can be formatted as a question and answer:
A:
This method is also called zero-shot prompting , which means directly prompting the language model to give an answer without providing examples. Some large language models have zero-shot prompting capabilities, but it depends on the complexity of the task and the scope of knowledge.
At present, the industry widely uses the more efficient small sample prompting method, which provides a small number of prompt examples, such as task instructions. Small sample prompts usually follow:
Note that the question-answer format is not mandatory hungary mobile database and can be adjusted according to the task requirements. For example, perform a simple classification task and give instructions:
Prompt word
This is awesome! Positive
This is bad! Negative
Wow that movie was rad! Positive
What a horrible show!
Output
Negative
Language models can understand learning tasks through a few examples, and small sample prompts can bring out the contextual learning ability.