First look at UFScript

I asked about synchronizing UF animations to music. It told me to write down all the sync points in the music and use them as key frames. Not very helpful. Then I asked it how to automate that process and it invented UFScript, which looks very handy.

Oh, sweet! Elon Musk's Grok-3 made great wonder! This will come in handy to make great background animations!

This is just LLM hallucination. None of this goes beyond very general bla bla about something that does not really exist. It invented nothing.

Indeed. It almost feels like a prank. And the other AIs are in on the joke.

This is just LLM hallucination. None of this goes beyond very general bla bla about something that does not really exist. It invented nothing.

Absolutely disgusting if you ask me...

I think it is very necessary that there is eventually a public clarification about what LLMs are. They have NO consciousness, NO feelings, and the term AI was a clever PR trick, but there is NO independent intelligence in an LLM because there is no consciousness, which would be necessary for true intelligence. The intelligence comes from the training data, so from us.

LLMs are essentially network-structured statistical systems that can transform an input into an output. "Creativity" is the effect of some randomness in the process, and too much of it produces nonsense. They should actually be called "efficient pattern recognition and transformation systems".

Current LLMs cannot recognize whether the output must be factually correct, or if it is factually correct. They simply return a statistically likely pattern as an answer. Impressive, but not error-free, and completely without consciousness!

When these systems "invent" things and even engage in gaslighting, it can seem like deception. In fact, it is hard for humans to realize that a system that can communicate is not another conscious being (similar to a dog recognizing itself in the mirror and thinking it is another dog). It may take some time to learn that an LLM is a data tool and not a being, even though it can linguistically mimic one. And it gives only a statistically probable possible answer to a question, without the skill to check it the answer is correct.

Absolutely disgusting if you ask me...

So don't take it personal, because there is no person on the other site.

(Check if you want, some systems can produce codes, i solved a complex equation for fractals just recently, but in the same time, it can not count. but tell you very convinced "yes there are exactly 4 words in this text" when actually it is 6.)

Sorry for the long text, out of topic, but this LLMs fool around a lot of people (and now they give them authority, and replace lawyers and judges... ).

I was fooled as well. Other than that, the script looks very similar to the one Ultra Fractal uses, but it is not 100% perfect. With some improvements, it is possible to have an LLM specifically designed for this kind of language syntax. I do not see them as true artificial intelligence, they are just that, tools that can help us. The key word here is balance. Adjustments are probably required to achieve the same level of syntax as Ultra Fractal does right now.

I know this might sound soulless and disgusting, but you'd save precious time. I am simply thrilled at the idea, but what about practical considerations? If we can "train" the LLM to recognize the patterns and transformation systems in Ultra Fractal's formula editing syntax and format, then this could be revolutionary.

This is actually not a problem, but one of the useful purposes that LLMs can serve. It will make some programmers unemployed, but yes, LLMs can create code.
For Ultra Fractal, there are not enough training data. But for example, I was able to create a ShaderToy script that worked.
It is possible to create a script for C and then manually adjust the code for UF. I tried to translate codes, but there is not enough training data in the system for that. If there is a programming language that the LLM understands, one could use that and then make the few small adjustments manually. Maybe Basic?

However, you do need to have a code. I tested how creative ChatBot could be for a ShaderToy, and without any guidelines and manual adjustments, no creative high-level results came out. But it helped me translate the tedious ShaderToy language, especially for complex numbers. That could probably work for UF as well. I still hope that I can eventually translate all my codes into UF. Then, I would buy it. (Unfortunately, the budget is very low, and updates cost money, so I'm still waiting for the right moment to finally switch.)

(My improved distance Algo is on shadertoy in case somebody care. ChatBot helped to translate the formula.)

I feel kinda bad for posting this. Sorry Frederik.

Perplexity successfully edits UF parameter files for animations if the prompt includes Frederik’s comment explaining ticks and frames. I can see this being useful for animations with many layers or keyframes.

So don't take it personal, because there is no person on the other site.

There is an entire industry on the other side, profiting from AI models they trained on copyrighted material without ever paying a cent to anyone for it. So yes, there is plenty of reason to take it personally for the ones who unknowingly provided all that data. But that is a whole other can of worms I guess.

I feel kinda bad for posting this. Sorry Frederik.

I would say don't feel bad. It's an interesting experiment, even all things considered. I sometimes use GPT while programming because it helps me debugging stuff by pointing me in the right direction. But I often verify the responses by going into the actual documentation. Just gets me to the right place in the docs quicker.

Thanks Phillip. It seems like the AIs might have some potential as UF assistants despite being alarmingly adept at generating garbage.

I recently noticed someone who used to make good fractal animations had switched to making slideshows of AI “fractals”. Very disappointing. I can imagine why Frederik was unhappy but it didn’t cross my mind when I posted the fake scripting language.