6 Comments

Excellent discussion as always, but I would argue that this shift has happened times and again, although here in a particularly salient form.

What I mean is that new technologies have rendered artisans obsolete, pushing for them to find meaning, and move to artistry.

Photo rendered obsolete portrait painters as a representation tool, is concomitant to an plethora of new painting movements that found meaning not in the technical resemblance of representation, but in the mood behind the strokes (Impressionism), or the decomposition of natural shapes in artificial shapes (cubism).

A tool rendering obsolete a human practice displaces it toward meaning, experiments and value addition.

This might be the beginning of what we are seeing there : what up to know was hard craftsmanship becomes mundane, and now an entire pan of craftsmanship is about to reinvent itself around new meaning.

Expand full comment

Life will change and events will be influenced by artificial intelligence. Are we losing our humanity or enhancing it? Only time will tell.

Expand full comment

I published a book of Hungarian poems written by GPT-2. (It was my COVID project in early 2020).

It did make a few poets freak out, from what I heard, but generative neural networks are no danger to creative professionals overall.

These are prototypes.

Expand full comment

I'm not sure I'd say "no danger", especially with how fast things are moving. You did that with GPT-2 and GPT-4 is likely to come out in a month or two, perhaps being as big of a leap up from 3 as 3 was from 2. MidJourney 4 is already pretty impressive and 5 is coming out Dec or Jan and supposed to almost always do hands well. Most people didn't even hear about all this until this year.

Expand full comment

Yes, but amazing stuff is always subject to selection bias, that hasn't changed.

The returns are diminishing, and the silicon behind the whale projects that can still deliver leaps in results is basically 5 years old by now.

GPT-2 can't beat GPT-4 for sure, but the former can run on a V100 GPU, the latter needs 7 figure cloud computing to be trained. Hell, I could train GPT-2 on my 13 year old Mac Pro.

I think neural networks have a lot of room to grow in practical, everyday use, but these will be very specific ones that can run on your iPhone's SOC. Those kinds have a decade of boom ahead.

I'm planning to write an effort post before the end of this year.

Expand full comment

It feels like protecting the prompt is more appropriate.

Expand full comment