Reality Check: Calibration Targets for Generative AI (AI Artbots)


Being so close to the generative AI space, sometimes you get vertigo. It just seems like the milestone advances, which used to break through every decade or so, are coming daily… on some days, even hourly. You get the feeling that something… something has… happened. Something profound that took A.I. from a slow slog through the ages to… to a near vertical ascent of the cliff face in front of it.

It’s almost as if we’re witnessing an… ascension.

Yet I also feel that being so close, I am capable of losing perspective; of getting caught up in the hype cycle… of confusing my hopes and dreams and projections with the actual present reality.

I only re-entered the world of “tech” in September of this year (2022, that is… more details on my 5-year zero-tech sabbatical later), not even 90 days ago, and it seems that in that very short time, progress has been made in quantum leaps. Could this really be true, or am I compressing “catching up with 5 years of tech growth” into 3 months, and merely projecting the illusion of hypersonic change & growth?

There is a possible answer to this dilemma:

check the data:

Code branches for Generative AI (StableDiffusion)

data supports perception: AI.art codebase rate of adoption & ascent is orders of magnitude faster & steeper than all other previous records (• star history of StableDiffusion AI.art codebase on github)

So, on one axis only, mind you, I present some “calibration targets” for generative AI engines, aka AI.art creation apps. That is, standard queries that I have asked the MidJourney AI to render for me, across the last 90 days. IMNSHO, the thing is learning at an insane rate. But why trust my judgement? Just take a gander at the generated images, and you can discern for yourself.

What follows are renders from MidJourney, using identical prompts, as rendered in September, October, and November of 2022… in other words, across a span of about 60 days.

from Zero to 1,000mph… in 60 days

So, how has generative AI “advanced” in the past 2 months? We present 4 reference “target prompts” to the MidJourney engine. The prompts were kept intentionally simple:

  • • eye,
  • • house,
  • • two hands,
  • • robot hand.

Results:

Nope.

I wasn’t wrong.

It’s eyes are opened.
It can see.

and It’s learning
At lightspeed.

 

, , ,