[This post was written by Leanpub co-founder Len Epp and published on Tuesday, February 14, 2023.]
Recently at Leanpub, we announced we are now supporting GPT-3 integration in our web browser book writing editor.
If you haven’t heard of GPT-3, it’s a powerful AI tool built by OpenAI that generates sequences of words based on prompts you provide to it. You can read about it here on Wikipedia and find more technical information about it on OpenAI’s website here, and read a great article about it by Stephen Wolfram here.
You can also watch me and Peter talk about it briefly in a Lean Publishing Podcast video here:
In this short post, I’m going to share some thoughts on how we think Leanpub authors might benefit from using this new tool in their writing processes.
Deciding What to Write (Next)
I’m am outliner myself, but a lot of people love to just dive in and start writing.
If you’re a dive-in writer, using an AI tool that responds to prompts could be a great way to think about what to write next.
Let’s imagine you want to write a book about the history of computing. You could start with something like this:
There’s no perfect place to start a history of computing. Do you start with the creation of the first machine, where an effect was designed to follow from a cause? Do you start with the creation of mathematics, where symbols and conventional systems were invented to carry out calculations? Do you start with [Ada Lovelace](https://en.wikipedia.org/wiki/Ada_Lovelace) and [Charles Babbage](https://en.wikipedia.org/wiki/Charles_Babbage), or maybe with [Alan Turing’s 1936 paper on computable numbers](/content/files/~robins/turing_paper_1936.pdf), from which we got the idea of a Turing machine?
I’ll paste that example into my test GPT-3 book on Leanpub and type two !! underneath it, to see what this prompts GPT-3 to produce:
That got me a pretty unsatisfactory result:
So let’s try prompting it directly, asking, “What should I write about first?”
That got me a pretty good result:
What this made me think was that explaining a Turing machine would be a great place to start, say for example in an introduction, before jumping back in history. That way, as you tell the story, readers can see the later relevance of each step people made towards the development of computers throughout history, even though the pre-19th century people I’d be writing about obviously had no definite idea where things were going to end up eventually.
The next step then would be to just try this: “Describe the main idea behind Alan Turing’s 1936 paper on computable numbers.”
That will get more text that you will definitely want to curate to make sure it’s accurate, but if you’re already an expert in the area you’re writing about, or you’re learning about it as you go, either way you’ve got a good “next thing” to write about.
Beating Writer’s Block
Writer’s block can take various forms. While the most severe form might be this kind (yes, that’s a very predictable clip from The Shining), one of the more common types is just an ordinary lack of motivation or boredom.
Just writing and editing some more or less arbitrary words can often be enough to get you going again.
Let’s try a straightforward enough example in Leanpub’s browser editor:
Here’s what I got:
How about we try something more specific:
Well, what do you know, I now have some good options for writing an entertaining aside in my chapter:
Now, you may be wondering, couldn’t I have just Googled that, instead of using GPT-3?
Well, yes, but the advantage of having an AI prompt in your writing tool itself is that it lets you maintain your focus on your writing, which is not really a trivial thing in our era of endless online distractions and rabbit holes.
What you might want to do in this case is note that in the future you could write an aside here:
Of course, when you go to write the aside at some point, you may find out you can’t rely on the AI prompt:
But still, you’d have the feeling of progress having just typed something, and planned an aside, which might be just enough to get you going writing again, or at least feeling productive instead of anxious, which is great in itself.
Rote or Programmatic Sections of Text
Lots of things we write not only are, but should be, rote or programmatic.
The most classic example might be an ordinary weather report: typically you will want these to be standardized output based on standardized input.
Another example is basic business deal reporting. Every merger or acquisition announcement, for example, will have some shared characteristics, like:
Company A, a something something company, bought N% of Company B, a something something company. The shares of Company B were purchased for $X per share, for a total consideration of $Y.
Let’s see how our AI assistant does in the browser editor, choosing a real-life transaction:
These are just a few pretty basic examples of how we think authors might use GPT-3 going forward.
If you’re a Leanpub author and you’re trying this out, please share your experiences on our Authors Forum! We’d all love to see what you’re trying, and how it’s working out.
PS In case you’re worried that AIs are going to take over writing from us humans any time soon, check out the Stephen Wolfram article I linked to above, that explains what’s really going on with all this. Here’s a screenshot:
PPS Here’s a recent book launch video with a Leanpub author who’s using an AI tool to draft his book: