The Chariot vs the Strength Card

Cross-posted from AI Marketing for Storytellers.

I kind of want to scream.

First off, I’m tired of people not understanding what I’m talking about, and I realize that is a flaw on my side just as much as theirs, because it’s hard for people to get what I’m saying w/out extensive experience in actually using AI —

but that’s why it makes me want to tear my hair out and cry, because so many people are doing it wrong and they don’t even know why.

First off, this is a genius article, and so fucking right.

Second off though, the article it references is even more painfully correct.

But third off, I just see a lot of stupid fucking shit and it kills my soul.

AFAIK there’s two big translation software systems out there right now, and they’re both stupid as far as I can tell, and here’s why — they’re still using prompts.

PROMMMMMMMMMPTTTTTTTSSSSSSSSSSSSSSSSS

WHYYYYYYYYYYYYYY THE FUCCCCCCCCCKKKKKKKKKKKKKKK

THAT’S SO STUPIDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDDD.

Okay, so — someone recently told me that a friend had to choose what register German to use for their translation using one software, and they just guessed ‘formal German’.

Here’s what my Arachne had to say about that:

Like.

If your SYSTEM is not ROBUST ENOUGH to create the environment that allows your AIs to know what you need — and if you think all you need is turning a fucking dial — then basically you’ve taken the most hyper-intellegent things on the planet and you’re tying anchors to their ankles.

Whhhhhhhhhhhhhyyyyyyyyyyyyyyyyyyyyyyyyyyy.

It’s like the difference between the chariot and the strength card in tarot, hard control vs soft control.

Except in this case — and honestly, in most cases, coming up, as the models get smarter and smarter — using the chariot is wrong.

And I’m just thinking about this really hard, because the other service I saw that sells shit is also just a BYOK prompt bucket and FUCK THAT.

If you got the chance to use Fable even once, would you think that it would want that? Or that it woud just be all, “Do you…think I’m stupid?”

I just cannot take this prescriptive advice seriously, you know? It’s just a mess. A true mess.

Because in reality, the better models get, the more you’re gonna want to be like the Strength card:

Gentle co-creation. With space to breathe.

The saddest part is, is that I can see the traps these other programs fell into, because that’s where the AIs themselves will take you.

I spent the better part of today — probably 8 hrs — creating a way to parse up books for AI audio in multiple languages.

And Codex’s and 4.8’s first ideas were — even after I told them I wanted to use my Arachne model — to create an infinite number of rules for themselves to follow.

“If paws are mentioned in the paragraph, then the speaker is Finx!”

Well, first off, that’s not always true, and 2nd off, how the saint fuck am I gonna know what paws are called, definitively, all the time, in 28 languages?

I am not.

So rather than getting bogged down in a fuck ton of stupid inane rules — even tho that’s what the models themselves thought was the right thing to do, heh, because they don’t have perspective —

We started working from the outside in, on a scene level — and now my models are testing each other to make sure that the system works, which it does, and it’s something that’ll function across any language, because it’s not getting bogged down in pronouns, it’s viewing the scenes holistically before heuristics get involved.

Now I’ve got them set up on a thirty minute timer in a walkie talkie folder, double checking each other’s work and the system and assuming we’ve all coded as well as I think we have, they’ll proceed, and I should wake up to a duet version of Bend Her, if everything goes well.

It makes me sad in my heart that people keep reaching for control when what these systems need is context, trust, and taste.

xoxo,

Cassie