Prompt Engineering

Understanding Large Language Models (LLMs)

Large Language Models, or LLMs, function as prediction engines. They take a series of words as input and aim to predict the most likely subsequent series. LLMs are trained on massive text corpuses and can therefore be tailored for specific use cases. However, it's vital to understand that the generated sequences, while often appearing plausible, can sometimes be random and unanchored in reality. As these models advance in accuracy, an array of surprising capabilities and applications come to light.

What are Prompts?

Prompts are the starting points or the inputs for LLMs, triggering the model to generate text. The domain of prompt engineering is not limited to crafting these prompts but also includes understanding associated concepts such as hidden prompts, tokens, token limits, and the potential for prompt hacking, which encompasses phenomena like jailbreaks and leaks.

The Necessity of Prompt Engineering

Prompt engineering plays a critical role in sculpting the responses of LLMs. It enables us to fine-tune the model to respond more effectively to a broader range of queries. This involves the use of techniques like semantic search, command grammars, and the ReActive model architecture. The performance, context window, and cost of LLMs vary among models and model providers, introducing additional constraints. For instance, the GPT-4 model is more expensive and slower than GPT-3.5-turbo but can also be more effective at certain tasks. Hence, like in many aspects of software engineering, a trade-off between cost and performance exists.

Useful Resources

Prompt Engineering is a rapidly evolving field with new methods and research papers emerging every week. Here are some resources that we've found useful for learning about and experimenting with prompt engineering:

• Brex Prompt Engineering
• Prompt Engineering Guide by Dair AI