Holodeck

I come from the last analog generation. Magnetic and mechanical data storage holds a certain fascination for me.

And I know what the pencil and cassette thing is all about.

And at the same time, I’m living in an era where, even without AI, the progress of the last 40 years is completely mind-boggling. Today’s phones used to be mainframe computers. Knowledge is abundantly available at any time and in any place. Digitization enables collaboration on a scale and at a speed that were once unimaginable.

Over time, I’ve taught myself various things whenever I wanted to do something but didn’t yet know how. There was no other way—and I think that’s a good thing. It makes life more interesting.

But since AI has become widely available, we have an interesting new possibility: We can do things that we don’t need to understand at all in order to do them.

It’s a bit like the holodeck of the NCC 1751-D USS Enterprise: We say what we want—and it appears before our eyes.

Essentially, I can now place an AI as a source of possibilities and methods alongside the internet as a source of knowledge and information—and learn from it.

I’ve attached the proof below. A Pong game where you can play against a very simple neural network. The network doesn’t know what to do at first—but it gets rewarded for successful behavior.
A genetic algorithm acts on this network, subjecting the neural network to evolution, at the end of which it’s still very simple—but will beat you.

I had no idea about neural networks or genetic algorithms. I still don’t—but I can now imagine something about them and experiment with them—because an AI spares me the necessity of learning things that I need neither in everyday life nor for earning a living, but which are prerequisites for such an experiment.

I now have a development environment for WordPress plugins, so I could use it along with its sub-agents and presets (also a learning process).

The initial task was:

Program a Gutenberg block¹ that illustrates neural networks and genetic algorithms. A demonstrator. I suggest that a network plays Pong against a human player and undergoes evolution through genetic algorithms depending on success.
Make a plan and discuss it with me. After approval, you will work on the task with your sub-agents until the degree of fulfillment measured against the plan is 100%. You work autonomously—token limit 5M. Ask questions at the beginning—otherwise only if you cannot complete your task without answers. Use high-quality sources for missing knowledge. No speculation allowed.

What can I say. It worked. This also includes an extensive system prompt, a context manager, various guidelines… But I’ll spare us those.

The first result was available and functional after just 25 minutes.
The debugging took a while longer. I also let the AI do that, given my lack of in-depth knowledge.

Now I could try out what this and that does with neural networks and how genetic algorithms can work by making changes to the program or parameters.
I still don’t “know” much about it… But I’ve gained a more concrete understanding of it.

1. A component I can use in WordPress to insert into this post [↩]