Using GPT has shown me that there are three broad modes of usage. Visualizing them as vectors, with both direction and magnitude, can be helpful in framing a solution to a problem. (To be clear, I'm referring to vectors in the simple directional sense, not in the sense of something like embeddings.)
The first mode is Generation. This is the most commonly discussed mode, where you can go from a short prompt to a long output generated by GPT. For example, providing a prompt such as "Write an essay about bears" will produce lots of text. This vector can be imagined as pointing "up".
The second mode is Transformation. This is an incredibly powerful mode, where you can take an existing artifact as a prompt and use GPT to transform it "in place". An example of this would be "Rewrite the following to use more elevated language" or "Rewrite the following using the style of H.P. Lovecraft". Both the prompt and the output are of similar length in this mode, and it can be thought of as pointing "sideways".
The third mode is Condensation. This mode takes long text as a prompt and produces a short output of some kind. Examples of this would be "Summarize the following into three bullet points" or "Give me five keywords for the following article". This vector can be thought of as pointing "down".
The magnitude of each vector is also important to consider. This is a way of understanding the strength of the application of the mode to the prompt. For example, "Write a paragraph about bears" is a shorter vector than "Write a 1000-word essay about bears". Similarly, "Rewrite this article to correct errors and flow better" is a shorter vector than "Rewrite this article in the style of Shakespeare".
By understanding the different modes and their applications, you can gain a more analytical way to think about your problem space, and make prompting feel less arbitrary.
Top comments (0)