Effective prompting takes a willingness to experiment with different request formats. One of the key aspects is working out which format or technique fits a particular task best — the zero-shot prompting technique, for example.
Zero-shot prompting is a method where the model is given a task without any prior examples or training. It is especially effective for simple tasks or when you want to test the model's baseline capabilities. Keep in mind that results can vary depending on the complexity of the task and the quality of the prompt itself. Here is an example of this prompting format used with GPT4:
Prompt:
Describe AI
Result:
Artificial intelligence (AI) is a field of computer science concerned with building systems capable of performing tasks that require human intelligence. This can include learning, speech recognition, decision-making, problem solving and natural language understanding. AI can be implemented in various forms, including machine learning, neural networks and deep learning algorithms.
The main types of AI include:
1. Narrow AI (ANI): specialised systems that perform specific tasks, such as face recognition or playing chess.
2. General AI (AGI): a hypothetical level of AI able to perform any intellectual task a human can perform.
3. Superintelligence (ASI): a level of AI that surpasses human intelligence in every respect.AI is applied across many fields, including medicine, finance, transport, entertainment and many others.
Depending on your requirements, you need different approaches, such as specifying the answer format or adding extra contextual detail:
Prompt:
Describe AI in three words
Result:
Smart computing systems.
The AI description example shows how refining the task can change the result. By adding the phrase "in three words" we gave the model a clear instruction that led to the result we wanted. Modern AI models have broad capabilities, from data analysis to composing music, but their effectiveness depends directly on the quality and precision of the input. Vague or unspecific requests can produce results that fall short of expectations.
Prompt engineering is exactly about composing optimal requests for different AI models. We will look at prompting techniques in more detail in section 2, "Techniques for building requests for AI models".