BTW all AI have an exponential S-shaped curve over all they can know, eventually all curves will "learn all" and become dead straight lines. Not sure about the start of the curve now. Yes a nanobot planet can "grow" and eat more "data" by expanding its homesystem, I'm not sure if that is linear or exponential I did once find that a growing civilization grows by 2, 4, 8....if you add matter to the outside they'd collect. How can growing after becoming the highest technology possible be a straight horizontal line, isn't it exponential or linear at least? Doubling your size doesn't take a trillion years and you protect your existence by double enough to last 2 trillion years, the probability of survival death is lowering by a actor of 2 and that factor is growing exponentially. Further, evolution has been exponential so far on Earth.
Returning to the start of the curve/line, at least for models, i believe it is exponentially less beneficial at any point of the curve for models, because there is trillions of rare patterns, only few are common, so your AI quickly gets good at a large chunk of the evaluation. The reason I believe now it doesn't get exponentially more data from semantic relationships is because most patterns are long prompts and rare, there are few short prompts/strings, so what are you going to say hey cat is like dog .... and 'fix the typo in the 8th word then add in the middle the word hehe' (one two three four five six seven eigheheht nine) is like 'um....me can't find a semantic for his it too rare'...nope. Back to universe, is a evolution seed exponential? Our history suggests so, and so does my prediction of the future immortality unit. If so, why are models dis-exponential!? In models the key is not data really or speed but insight extraction e-mergence. Yes in real life patterns, it is scale that is the powerhorse. WTH is this. BTW I've made a dozen large discoveries this last months and will soon share something mind-shatteringly big for free. The model, is what generates new patterns, it's not data based directly but rather based on learning a hierarchy of pattern types, and heterarchy, hmm, it learns the patterns of patterns, it generates exactly what to do, scale only allows death of nodes to be made up for. The model is what efficiently finds the patterns, there is only so much data it needs to do this it needn't eat a galaxy full like brute force, so while it seems to not be data based it is just pattern based... It's not about adding different data, it's about making patterns and learning the hierarchy of patterns, not every bit and string! The scale of real world homeworld is scale not pattern because at some distance a unit cannot speak to other units and so you get some individualness to the giant galaxy sized megastructure perhaps, it does form patterns but this is in accordance to the model, at some point it must be just scale. And it must be, because if it weren't cloned units then you'd have a non-robust system, you don't want there to be only 1 telescope, nor 55, but 1 per each unit, and nor that each has to talk to each to benefit, I mean its most dependent on its neighbors, so if a signal must be sent from one end to the other it can just do the same thing at each unit in between... ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/Tfce80000509c1fb3-Macf52db2c2868b2a9ef9d07c Delivery options: https://agi.topicbox.com/groups/agi/subscription
