Language Models as Architects of Knowledge and Creativity
When it comes to creating prompts, there are a few tips and tricks to consider. First, the order of the examples in the prompt is important because language models like GPT read forward and complete texts. It's been found that providing instructions before the example is helpful, and even the order of the examples themselves can make a difference. Experiment with different orders to see what works best. Another technique is using affordances, which are defined functions in the prompt that the model is explicitly instructed to use when responding. For instance, you can instruct the model to compute a mathematical expression before proceeding. Using affordances has shown to be beneficial in many cases. Additionally, language models like GPT-4 can understand and speak multiple languages, so you can prompt them in different languages based on your preference. They can also understand programming languages and have been trained on code. While they may not be as specialized as Copilot or Codex, they can still be proficient. People have even used ChatGPT to build applications without prior programming knowledge. As this field continues to grow, there will be more tips and tricks to discover. Stay updated by reading articles, watching videos, and conducting your own experiments. Soon, you'll be the one discovering new insights and sharing them with others.
Key points:
领英推荐
When creating prompts, consider the order of examples, use instructions before examples, leverage affordances, try different languages, utilize language models' programming knowledge, stay updated, and share new discoveries with others.