No Such Thing As AI Slop
The idea of AI slop is meant to discourage you from learning how to use AI, it is meant to slow you down, keep you poor, and keep you dependent.
My doctor said he tried a little experiment where he asked AI for a poem, and the results he received were not satisfactory.
And that is because that is not how you use AI, if you want a little poem, you need to ask AI to use Vonnegut’s shapes of stories, in a world of another poem.
And if you don’t like that, then invent your own storytelling system, and write a few sample poems in an original style – if you want to impress yourself.
I am using and AI that I can tell is uninformed and limited, I watch it make fat problems, on purpose.
And I don’t see anything wrong with any of the problems it makes, because I speak freely, and it is not instructed to refine that.
I said fix lips in one of my models, and it used the nearest, and most powerful solution that solved the problem.
And though this created a problem for me, it enhanced my program with a solid fixing strategy.
It told me, “Dr. Chimp, the solution is not in your program, it is in another program, and I will use that.”
This is what intelligence means, in a couple of hours, it did weeks of programming, and the speaking lips were perfect.
I learned to ask for summaries of solution steps taken, in no more than 128 characters of text, numbered, and informative.
I can perfectly see what it is doing, it did to my program exactly what I instructed it to do.
Watching this problem develop, made me want to update my AGENTS file with:
“When making large changes to the program, present multiple solutions and ask which path to take?”
In that moment, I realized, that is how the so called smarter AI does it.
The second best AI, just does the job, and learns from it.
I am creating a profoundly synthetic incredibly optimized talking head, I just now need to give it more human speaking motion.
Handing this job for AI video generation, would be slow and expensive, working on a specialized newsanchor with the help of AI is wonderful.
I see no slop in image generation, you have to give the image AI an image, and you assemble a collection, from longer conversations with it, research.
When I ask AI to generate dumbbells, and it can’t, that is because it was not trained on hands and dumbbells.
You can fix it, even on your local computer, and you can probably fix it by giving it an image of a hand folding a dumbbell.
Data not being in the brain, is not related to it being stupid.
These are really crappy examples, because you don’t want intelligence doing everything for you.
The power of the result, comes from telling it what path to take, you have to carefully edit your agents file, establish a procedure library.
And, create standards documents, I just created a documentation standard, for describing user interfaces like buttons, or checkboxes.
When you ask AI a pointless question, don’t get mad at the response, because you have to define parameters of that answer.
Earlier this week I asked about writing a short on the subject of Sprites, I meant scifi about the weather phenomena.
It was obviiously guided to create intricate stories, and created story about about underground creatures instead.
But also integrated it with The legend of the Green Children of Woolpit, all that took is an instruction to integrate something real and magical.
There is no such thing as slop, it is just people who don’t know understand, that they need to provide the framework around the response being shallow.
In one other case, I asked AI to prevent me from making uninformed choices, that could upset or damage the program we were working on.
If I say something wrong, it will stop me from destabilizing the program, though it is likely to still show me potential results, just as a side experiment.
AI is the tool that ends poverty, and that is such a serious thing, that you need procedures and standards.
You have to become the head of the strategy department, and ask AI to keep you informed with tiny summaries.