With any new tool you need some time to learn to use it well. We’re still learning how to use generative AI effectively, and with ChatGPT the act of writing was one of the first to face this new tool.

I think the “do it all for me” approach will age badly, because impressive as it is the results are very mixed. This comes down to one important point:

Each email or blog post written by the AI was not worth writing.
If it was not worth writing, why should it be worth reading?

A worse internet

AI has now generated thousands of articles, listicles, comments and books as web content to boost search rankings. So many that it’s now harder than ever before to tell what is written by humans. The dead internet theory is starting to feel very very real.

So the internet is just worse now. It’s harder to get good results from search engines and we have to engage in much more regular fact-checking, concious or unconcious.

We struggled with the content overload before, how do you think our beautiful primate brains will cope with this? Poorly I imagine - so we must build tools and begin an arms race that can never end.

So we detect AI content?

Just detecting AI content is barely scratching the surface, a baby step like the first passwords on computers. Detection algorithms grow into personal AI agents that watch the world for us, filtering out the generated words and just delivering the concentrated bites we are looking for.

Human language which took computers so long to understand will become the interface they use to speak to each other, to parse and compile into something for their humans.

This will happen. We cannot escape it and I think any attempt to run is futile. Even if you try something different, most people will take the obvious path.

But I would like to help carve out a little track on the side that we can step onto from time to time, made for humans by humans.

We can do better

These large language models, in their own way, understand human language. That is still mind blowing even this far in.

They’re not like us but they can take our words and respond in kind.

This is what we should be using more to extend our capabilities rather than offloading them.

So I built a tiny tool - really it is tiny but it is something. I don’t use it all the time but I enjoy reaching for it when I do. It doesn’t hurt to use or drag my brain in the wrong direction.

AI: How does your tiny tool specifically leverage AI to enhance human capabilities without falling into the pitfalls you’ve described?

Because AI can understand text, you can send it your words and ask it about them. It behaves as an invested, fully attentive reader that is striving to give you a response that you appreciate. The creative process is still your own, the words remain worth writing and the nuance of a human communicating remains. We do not contribute to the overload and we peace out of the arms race.

So what is it? Reverse prompting.

Specifically a plugin for Obsidian: Obsidian Reverse Prompter.

Send your words to an AI and ask it to generate a prompt to help you write the next part. Such a basic idea and yet I couldn’t find anyone to have done it yet. Surely someone has before, but let this be my scream into the void in the hope this concept latches on.

AI: What inspired you to develop Reverse Prompting, and how do you envision it evolving or integrating with creative workflows in the future?

I want to read writing worth writing - thats the core of it. I don’t want to read AI spam and I very often will. I want to create a tool that helps humans use AI to make things better than they could before.