Fail, fail, fail, fail, succeed

Waterfront at Night (Part 2)

We are living in a transitional period in human evolution; more specifically, the phase in AI development where it begins to appear everywhere, yet, for the general public, it seems almost transparent. Like it’s not really that big a deal.

But it is a big deal. One AI researchers already don’t fully understand. Case in point: DALL-E 2.

Yesterday’s image “Waterfront at Night,” was created from a short text prompt by the Generative Pre-trained Transformer (GPT) model initially developed by OpenAI in 2018.

After getting access to the free program, this was my 8th attempt to have the algorithm generate something cool. It’s so simple a child learning to read could use it.

Because it uses machine learning (I.e. it teaches itself by analyzing input, in this case over 6 million images), the designers can’t reverse-engineer how it actually creates a specific image.

In other words, humans created this thing and then gave it the ability to recursively teach itself – so unless it tells us, we don’t know exactly how it makes its decisions.

And yet here it is, in the wild, happily teaching itself how to become better at what it does. To OpenAI’s credit, they’ve tried to limit its capabilities (no porn or violence), but at the end of the day, good luck with that.

Welcome to the future. Ready or not, it’s here and raring to go.

Or perhaps I should say, it’s here and raring to show us what it can do.