When we last left our Abecedarium, with the word Judgment for the letter J, I decided to switch gears a bit. I hinted that I would find my “K” letter through the help of OpenAI’s ChatGPT. You probably live under a rock if you have not heard about the emergence of these “large language model” AI (artificial intelligence) programs. To explain, these LLM’s, as they are called, are programs that have read literally everything on the internet and can return “answers” to your “questions” in common prose. Your “question” may be a prompt like “explain to me how a human eye works” or “plan a two day vacation in Seattle” or “why might I get a strange rash between my toes”. The “answer” is developed by the LLM digging into all the text it has consumed and then building a response by predicting what the next word should be based on the previous one(s). In other words, LLM’s interpolate. They don’t exterpolate. Nor do they understand the difference between reliable information and total bullshit (see: question about that rash).
Right now, people seem most concerned about these LLM’s putting people out of work. I have no doubt that might happen. But tech advances have always obsoleted some jobs and yet created many, many more. The trick, in my opinion, is to figure out how you can use something like an LLM as a tool—not how you can fight it (see: current WGA strike). Here’s an example my analytical science friends will understand. Back in the prehistoric days of chromatography (a technique for separating compounds in a mixture), if you wanted to quantify the amounts of the compounds in a mixture you would do this: take the spectrum which is drawn by the instrument on a strip chart and cut out the peaks (with scissors!) associated with the different compounds; literally weigh those cutouts; then perform mathematical calculations involving those weights and response factors previously determined. Now (and by “now” I mean since the 1980’s), a computer program does all that for you, in addition to running the experiment. Did that put a lot of scientists out of work? That’s a LOT of labor that went away. No. It freed those scientists up to think about what the results meant in the context of why the experiment was run and to dream up more experiments. Scientific progress sped up dramatically.
I believe these LLM’s can be similar tools. Since they currently can only interpolate, they can’t do the really creative stuff. What they can do is help people like me that hyperventilate at the sight of a blank page. So. Let’s have ChatGPT help me write this essay. There was no obvious “K” noun that occurred to me as I wrapped up last time, so I put in the prompt “Generate a list of five random nouns that begin with the letter K”. The program returned these nouns: kite, key, kingdom, knife, kaleidoscope. So far, so good. Then I asked it to write a 300 word story using those nouns.
I quickly figured out the first new job role that will be created as a result of these LLM’s: Prompt Writer. Just as the people who performed, by hand, the laborious calculations to send the first astronauts into space later became the first programmers of computers that would do those calculations, I’m guessing writers will become the first professional Prompt Writers. My initial observation of ChatGPT’s response to my prompt is that the story was 419 words long. It’s not that ChatGPT is not that precise with math. It was that I was not precise with my prompt. The story did indeed include 300 words. But I wanted a story NO MORE than 300 words. Next, I noticed that by requesting a “story,” the tool returned what is likely the most common type of writing associated with that word: a children’s story. What followed was 419 awkward words about a faraway kingdom where a special little girl found a box that contained a kite, a key, and—wait for it—a magical knife that granted wishes. Of all the common objects I could think of that I might expect to grant wishes, a knife would NOT be an intuitive choice—unless the magical wish is to maim someone. Finally, I see this output as a children’s story that is not entirely different from many children’s stories that are out there—which is probably where it came from since the LLM “learned” from existing stories.
I tried again: “Write a paragraph no more than 200 words in length that includes the nouns kite, key, knife, kingdom, and kaleidoscope.” The response was another children’s fantasy story, this time a precise 185 words, but very similar to the previous attempt. One more time: “Write a non-fiction paragraph of no more than 200 words that includes the words kite, key, knife, kingdom, and kaleidoscope.”
The response: “In the world of art and craftsmanship, the creative process often resembles a kaleidoscope, where unique elements come together to form a harmonious masterpiece. Just as a skilled artist wields a paintbrush, a chef wields a knife with precision and finesse. But beyond the realms of traditional arts, creativity also finds its expression in the intricate designs of a lock, waiting to be unlocked with a key. In the field of literature, storytelling is akin to building a kingdom of words, where characters and plotlines weave a rich tapestry of imagination. And just as a kite soars high in the sky, propelled by the wind, ideas take flight in the vast expanse of the writer’s mind, driven by inspiration and passion. Whether it’s painting a vivid picture with words, crafting a delicious culinary creation, unlocking the mysteries of a story, or creating a world with words, these diverse expressions of creativity remind us of the kaleidoscope of possibilities that exist within each of us, waiting to be explored and shared with the world.”
Better, I guess, but sort of a word salad. A little freaky, though, that the topic it chose was “creativity” when I had just been saying that LLM’s aren’t really creative! The tool is only as good as the prompt, though. ChatGPT is trying to give me what I want. I tried one more scary thing. I asked it to write said non-fiction paragraph in the style of Sherri Bassner. It should have sourced all my blog entries, articles for Lab Manager, and scientific articles, then pulled together a short paragraph in the style it assumes is mine.
Here is the result: ” In the realm of human experience, creativity flutters like a butterfly, drifting from the vibrant kaleidoscope of one’s mind. It takes a delicate touch, like the feathered stroke of a painter’s brush or the precise slice of a chef’s knife, to unlock the possibilities within. Just as a key fits perfectly into a lock, creativity unlocks the doors to new worlds and uncharted territories. It is in these unexplored realms that we build our own kingdoms, shaping stories that breathe life into our existence. Like a kite soaring through the limitless sky, imagination takes flight, capturing the whimsical essence of our dreams. And just as a kaleidoscope refracts light into a symphony of colors, so too does creativity refract our experiences, transforming them into something extraordinary. It is through the lens of creativity that we transcend the mundane and discover the magic that lies within ourselves and the world around us. With the intertwining threads of the kite, key, knife, kingdom, and kaleidoscope, we unlock the hidden depths of our imagination and create a tapestry that reflects the beauty of the human spirit.”
Ugh. Please don’t tell me I write like that! Sure, I like a good metaphor but, really! Let’s just say I’m not particularly worried about ChatGPT obsoleting my blog. I also don’t think the Writer’s Guild should be too concerned. Sure, the LLM’s might take on a lot of the derivative writing. But the writing that really matters, the writing that means something and will pay well, has to be original—and the LLM’s can’t do that. Yet.