System Prompts and Prompting Types

System prompts are predefined instructions that is given to the model that governs the flow of output, defining the models of its role, any specific instructions if any while enforcing boundaries for its usage.

there are primarily 3 types of prompting -

1. Zero shot prompting - this is when model is given a direct question, without any primary examples - be it direct from browser in a new chat or a simple straight forward api call. Though, this is inherently fast, it can still lack in consistency and can sometimes provide vague responses.

2. Few shot prompting - this is when a model is given a direct question but with considerable amount of examples (3-4) for it’s reference. This helps in receiving structured responses, code - writing etc.

   To use this method to its potential, user must provide good examples with clear instructions.

3. Chain of thought - COT - this is when a user asks model explicitly to reason step-by-step before giving an answer. This is the model that is used by big EdTech/Software companies during AI integeration. This slims the chances of vague responses and helps in providing logical, planned and calculated responses