AI made outcome harder, but increased output
AI has increased amount of communication, in other words output, but not the outcome.1 It has decreased amount of ideas overall, density of notions per communication unit,2 and increased cognitive load for each knowledge worker.
As always, there’s no silver bullet, however we can work with these issues even now.
Density of communication
You probably don’t want to make AI’s text too dense. High density text can’t be easily understood. Try a philosophy book (e.g. Hegel:). If a text is too dense, there’s high chance you’ll lose ideas and connections between them.
At the same time, AI’s text may look too low on density3: you lose the idea while trying to understand it. My solution to it is to use proper instructions and experimentation. More on that in the next sections.
Amount of new ideas
Novelty of AI’s ideas is an answered question: AI does not produce new ideas yet. LLM works as a generator based on a huge amount of texts, so it finds something applicable in your context, but doesn’t create anything from scratch.
However, novelty is in the context. For one, an idea of Agile is a disruption of their whole identity, but for other folk same happens when they hear of USSR’s 5-year-plans. So AI can generate something new for your specific context.
You don’t expect AI to produce new market disruption for you, but use it as an extended mind — and chance to get valuable result gets higher.
What worked for me
While struggling with all the volume of text coming from AI, I have created this approach for me.
-
Visuals.
An LLM is not limited to text and can generate other forms of information: diagrams, tables, clickable visualizations. Use it for your purpose instead of a text.
I believe that information in form of a text (just words following each other) is not a cure. You may ask AI visualization instead of text so that all the output is visible at one sight. But I have already spoken about it here and here.
-
Create a skill that writes actually readable text.
As I have an understanding of what exact form of ACs I like (and find readable) or texts produced by myself (incl. this one) I can formulate what discourse characteristics I possess and teach LLM to re-write in the way that I find better looking.
So use your AI buddy with the proper density in the text, don’t expect it to create a new business for you, and set up proper rules.
-
There’s the Jevons paradox: when efficiency grows, usage grows with it. So you can see how easy it is to produce a text now. ↩
-
Written text does not transmit the whole message. AI fills the gap with the amount of text, while people solve this issue with communication: memes and overall context, hearing, seeing, touching. We aren’t limited to readable symbols. ↩
-
Density is not a characteristic that shows if a text is good or bad. You can see good texts with high density (e.g. Finite and Infinite Games by James P. Carse or Seneca’s Letters) and with low density (The Goal by Goldratt or The 80/20 Principle by Richard Koch). ↩
— Aliaksei