Can't be a computer because that generates a new homeostasis.

AI in some respects in this generative context is a rat tapping the feeder bar to get a food pellet, where eventually it sets the rules and buries itself in food. All models will have this—it's just a question of what the exponent is.

You cannot fix that sort of thing with more complexity!! The more complicated the feedback loops are, the more robust the homeostasis becomes. You can't even fully fix it by restarting the interaction, not without some human filtering out "here's the helpful bit."

It's that the AI text predictor as it runs is exponentially decaying into a steady state as it consumes more of what it produces. So it has seen itself dithering, it has seen me yell at it for dithering, so it **assumes** that that's the interaction it's supposed to be generating **more of**.

It wouldn't stop! Just endless going around in circles.

Now I actually used to work on these sorts of systems a very long time ago, Markov chains, and there's something that I guess AI researchers probably all know about but users probably don't, intensified by the complexity of the system:

She is let out and reveals that the secret is a secret song that charms foxes, which she sings and Pied-Pipers him directly into his own pixie-box, which she shuts tightly, and leaves him there until he dies.

The moral I guess is don't fuck with faeries?

But through raw persistence the fox finally has a day where the pixie slips and leaves the door unlocked; he shuts her in a box and laughs maniacally.

She knows the fox is too clever to resist a secret and so offers one in exchange for being let free, he agrees as he thinks he'll get her again.

Being wily, the fox attempts to trick the pixie out of the house: I know of a turnip field we could steal the turnips from, "too tired to steal turnips today, maybe tomorrow morning" only for the fox to find that overnight the turnips were pilfered (actually by the pixie) etc.

The big bad wolf was actually a sneaky fox! But he was indeed able to burrow into the wooden house and, with some effort, into the stone house. The pixies don't escape, they are just eaten. This sets up a strong villain character for the last defenseless pixie stuck in her little iron house.

Is the pastor patient, kind, reliable, self-controlled, calm? Do they celebrate others doing better than themselves with joy it react in envy? Do they avoid calling for war against the enemies of the church just as Christ said “forgive them they don't know what they are doing?”—and so on.

How do we keep ourselves safe, when the enemy is inside our walls?

We test the spirits, we test the doctrines, we test the roots. We never stop testing. We are spiritual scientists.

And how do we test? We look at the people that the doctrine produces.

In addition the 1st letter of John says to test every spirit, meaning angels/demons, to see if they are really angels or not.

The only conclusion is that Satan is in the church. Has been from early days. Your little Protestant denomination, probably is not safe from that either—Paul wasn't.

Proof of how quickly things can turn around that I am already eating those words XD

Struggling is an understatement. Mahomes trying for a first down rather than 3 points that last play? He's going to set history for the only team to get totally shut out at a Superbowl!

They all had Superbowl commercials that celebrated diversity and equity and inclusion, and then they all had cowardly CEOs barking orders until their marketers scrambling to come up with new ads at the last minute...

At least that's my headcanon

So like if I come up with a song on my guitar nobody is like “will you record an album?! how much money will it make!!”

And yet with writing...

I worked on a software team that consistently beat its deadlines and ultimately worked ourselves out of a job—the game was finished, it wasn't a great idea for its audience, the company wasn't a game studio, etc.

Been on three teams since; none is interested in remaking that magic.

But second and more importantly, you get fucks like me who need to convince you about the importance of summarizing every YouTube clip with our new app because we want those transcripts and we gotta justify the 3 months we sank into that GPT frontend and my personal usage is never gonna make it work

So these are the two real dangers of AI models, I guess: firstly they can just ignore half the data if you don't call them on their bullshit—think like HR reps using this to filter which candidates are good and it just drops some resumes, ignores the work experience of others. Lots of limitations…