AI and LLM Miscellany

Prompt Engineering, Exploration

I've spent a lot of time exploring the latest AI- and LLM-related tools on my own, as it's an area of interest and ongoing fascination to me, both as a designer and as a writer of fiction novels.

In fact, writing Book Blurb Pro—my own stand-alone GPT focused on book blurbs (available here)—allowed me to pursue several of these interests at once, in addition to book marketing. This is particular task is well-suited to an LLM, since online resources about how to craft an effective abound, and because of the strict structure book blurbs tend to adhere to.

Building a GPT, spelling out its directives

The key to creating an effective GPT—at least as far as Book Blurb Pro is concerned—was first to provide it with the best storehouse of knowledge and guidelines available, from experienced writers and reliable sources. Next, the GPT was designed to continually ask the writer questions, so that it could glean the information it needed over time, and guide the writer toward articulating key details about characters, setting, conflict, and plot.

The goal was to fashion a GPT that didn't rely on a one-shot, query/response model, but rather one that works with a writer to come up with a blurb in much the same way the writer would alone.

So many images…

I've also spent time using AI image generators—primarily Midjourney—to generate thousands of images. I've found that it's a great way to explore ideas. And it's been a great source of inspiration, largely because the way they draw on the underlying latent space results in unexpected attributes, from styling to framing to colors, even when these things are articulated in the prompt. The inherent variability of the output is just as compelling to me as the promise of being able to lock down (aka cohesion) key aspects of the imagery!

I hesitate to say how much time I've spent poking at these tools (toys?), but will instead allow the imagery to speak for itself.