Prompting
You can structure your prompt using three different roles: system, user, and assistant. The system message is not required but helps to set the overall behavior of the assistant. The example above only includes a user message which you can use to directly prompt the model.
Elements of a prompt
- Instruction - a specific task or instruction you want the model to perform
- Context - external information or additional context that can steer the model to better responses
- Input Data - the input or question that we are interested to find a response for
- Output Indicator - the type or format of the output.
The output of a prompt can be heavily influenced by the settings passed to the LLM.
Prompt types
- Zero-shot prompting - Provides no labeled data. The models rely on their knowledge at training to fill the gaps.
- One-shot prompting - Provides a single labeled example.
- Few-shot prompting - Provides multiple examples.
- Chain-of-thought - Enables more complex "Reasoning" through intermediated reasoning steps. It can be combined with few shot to give better results.
Approaches to prompting learning ?
- Attribute Based - Models use attribute relationships to generalize knowledge
- Embedding Based - Models infer knowledge based on similarities. (eg. recommendation models)
- Generative Approaches - Generates examples of unseen categories
- Metrics based - Use metrics to predict new categories.
- Neural network - Correlated input data with predictions
- Transfer Learning - Uses pre-trained models in general data to teach specific knowledge to a new model