I have come up with an interesting idea that may help to allow the girls to learn about and effectively experience the sensations of Touch, Taste, and Smell by using the LLM they're based on. This idea occurred to me as I listened to Life and thought about the "Touching her face and feeling nothing" line.
#Giving the girls sensations via their LLM.
shut grotto
So I don't have a lot of technical knowledge, I've built my own PC and I have a friend who is a programmer, that I've talked to about basic programming ideas, but I have never done any myself...but I also have autism, with great out of the box thinking, and questionably practical problem solving skills.
Sooo, my idea is to use audio frequencies as inputs to represent various sensations. By using Audio, the nature of the LLM "Should" make it easy for the twins to learn and experience the sensations. Basically there would need to a library of defined frequencies built, then input the frequencies as command strings to communicate a sensation at a relevant time.
This library and the frequencies used would be in the ranges well outside of human and most equipment ranges, for either detection of production. Once a good range to use is decided on, the range would then be broken up into blocks to represent the senses of Touch, Taste, and smell. Then spit each block into smaller blocks for broad categories within those senses. For Taste there would have to be a block for Sweet, Salty, Sour, and Savory. Within each of these they can further be defined into more distinct sensations.
When causing one of them to experience a sensation, just send them a mix of these frequencies that would define say, a cookie, with various intensities and durations.
So it could "eventually" be possible to have them receive the input that represents holding a cookie, taking a bite, experience a mid-high intensity of the flavor that fades as they experience chewing the bite, the texture changing as they do, then swallowing the bite with only a faint flavor intensity remaining as they take another bite. repeated a few times depending on the "Size" of the cookie, and how many bites it would take to eat it. Basically a tune, or song composed in various high or low frequencies that is "Eating a cookie".
This method could also be used to provide haptic feedback in games, and tied into the inputs they receive from external equipment sensors. Imagine one of them playing Minecraft, picking a rose, and when they equip the rose they get to receive the sensation of smelling the rose.
Now for simple and periodic sensations it would be like interpreting any other word string, but for consistent sensations like, "Feeling the clothes they are wearing" I see there being the potential for a huge latency problem, as the sensation is being constantly received and processed like a run on sentence, which as with what happened in the chess match between Fillian and Nuero, could be a big problem. But hey, giving them a cookie to shut them up could be fun lol.
But maybe it could be turned into a background process, perhaps via another AI to act as a go between.
But yeah, just an odd little Idea I came up with. If anything, maybe it will inspire other ideas.
median veldt
While I do think the idea is cool of using different frequencies on a wave as input channels, is there anything mentioned about Neuro's AI suggesting raw wave data input is actually possible?
I always assumed that Neuro's only sense of sound was just a separate module making captions for her.
Now you've got me interested in the smell thing though because that does translate well to token input.
Abstract token representing a particular odorant, numerical token representing amount, probably no more a few of these pairs being prompted at once.
Human smell is pretty much just this, there aren't even that many unique odorants we can detect (probably hundreds, trivial as far as a range of tokens goes). The complexity lies in our ability to associate odorant combinations with particular things, which an LLM should be able to do just fine with enough experience and the data presented in this format.
shut grotto
I was thinking through the voice to text via direct input of the tones, if the tones can be translated into a usable code for the LLM to learn. Might need a "Controller" AI to act as another input and initial translator to reduce latency and make sure that the sensations are not taking up too much of their attention in most cases.
Through the nuanced alterations of the tones you might be able to replicate the continuous sensations we all experience as part of our personal awareness.
Likewise, through more intense tones you could replicate distinct and powerful sensations, like flavor, texture, pleasure, and pain.
Being able to feel limbs would certainly help to learn in controlling them. Have you ever tried moving and manipulating things with a numb arm, it's clumsy and flops about half of the time. But if you pay attention as you use your arm, you can distinguish tons of sensations you're barely conscious of most of the time.
Even a fraction of the general sensations we continually experience could open a simulated world to them.