Daily, snackable writings and podcasts to spur changes in thinking.
Building a blueprint for a better brain by tinkering with the code.
The first illustrated book from Tinkered Thinking will soon be available.Subscribe below to get a notification.
July 22nd, 2020
Tinkered Thinking is developing a small simple app as an experiment to try and put GPT-3 to good use.
(I promise Tinkered Thinking is not going to go one and on about GPT-3 forever.)
As was mentioned in the previous episode about GPT-3, it’s pretty easy to imagine how it could be put to wicked use. That’s ‘wicked' in the New England sense of wicked cool, but also the New England sense from the 1690’s, which means: we should probably burn it at the stake. It’s a tool that’s sharp which can be both very useful and very dangerous depending on how we ply that edge.
Constraints are key to GPT-3 use. The prompt that it is given is essentially a constraint through what we might think of as context. The idea that Context is everything seems to hold pretty solidly when we examine the constructed perspective of something that is not human.
The app in development is an experiment on several levels. But first, what’s the app do? The concept is very simple: The right question changes the course of our thinking. The imagined use is for writers, artists and creatives of all types to enter a topic, question or short idea when they are feeling ‘stuck’. The Tinkered Question App will then use the magic of GPT-3 to generate 3 insightful questions that provoke our bogged-down creative to think about their artistic situation in a new way with a new perspective.
A great question doesn’t merely create a void where we should imagine an answer; a great question is an open-ended concept that creates forward momentum.
The question, as a concept and a tool also has one subtle aspect which functions like the flux-capacitor at the heart of it’s magic: a question changes the context of our thinking without explicitly introducing a new idea.
This is a sly hack that seems to have arisen within human thinking. We’re all familiar with the itchy experience of going to a friend for help and getting a bunch of suggestions that just don’t jive with our thinking. The experience seems to be symmetrical: it’s equally frustrating to give a friend a bunch of suggestions to their problem and see each great idea fall flat before their unimpressed psyche.
But an insightful question seems to have a unique ability to thread itself through this prickly gulf of incongruent perspectives. It’s the trojan horse of dialogue: the well formed question sneaks into the other person’s mind under the guise of their context, but then warps it, bends it.
The first experiment is this issue of context, and it’s not a puzzle in terms of coding but in terms of language. In order for GPT-3 to produce questions that are impressive, insightful, and ultimately helpful, then there needs to be a context larger than the user input for this to occur. This fascinating challenge might end up being the heart of apps that utilize GPT-3. The difficulty or ‘value’ of such an app might not be the coding -which many coders can easily accomplish- but with the unique language construct that is used to unlock GPT-3 in a very specific way, complete with tone, character, flavour and depth of insight. This challenge presents an interesting crossroads between the world of coding and the humanities. As powerful as GPT-3 is, it’s not a mindreader and it emphatically won’t do exactly what you want, nor plan. It’s a bit like another person in that you have to coax it into a certain mood and mode of thought with your own flavour of language.
The second experiment that lends well to this idea is that it can generate a massive number of questions, and if users enjoy the app, they will be able to highlight their favourite questions, thereby creating a filter for output which can than be used recursively to redesign the unique, hardcoded prompt for GPT-3, and on top of that, the database of questions lends itself well to research about the nature of questions themselves, which is a major source of interest for yours truly.
The third experiment, is obviously, the business potential behind an app. Such products have laughably manageable costs for getting up and running. The only downsides are really time spent actually building the app and tinkering with the hardcoded prompt and any other hoops that might be required to jump through. It’s a bit like a lottery ticket which costs only the amount of time it takes to fill out, but perhaps with better odds, who knows? The real question isn’t why build it? The real question is why not?
The shift between these two questions really occurred after witnessing McKay Wrigley build one of the first GPT-3 apps and publicly share it online, an app called LearnFromAny1 which allows you to enter a well-known person’s name and a topic and have returned an explanation of that topic in the manner of the person. The speed at which he was able to bring something to a stage where it could be demoed spoke for itself: the effort is just not that much of an investment and certainly presents no loss at all. It’s thanks to McKay that focus went from the questions why, or what, to why not?! And this is exactly the sort of shift in thinking that is at the heart of hope for this app: is it possible to give people a tool that helps them ask better questions?
If this app sounds interesting to you, and you think you’d like to be part of the beta-testing group once OpenAI gives the green light for that phase of the process, then please subscribe to Tinkered Thinking on the website. Beta-testers will be drawn exclusively from the subscriber group.
donating = loving
If you appreciate the work of Tinkered Thinking, please consider lending support. This platform can only continue and flourish with the support of readers and listeners like you.
Appreciation can be more than a feeling. Toss something in the jar if you find your thinking delightfully tinkered.