🧵 Untitled Thread
Anonymous at Wed, 16 Oct 2024 09:10:22 UTC No. 16434100
Consider this. Suppose I can create an algorithm that can create an analytics function for a finite sequence of numbers. The word finite here is very important as the simplest function that satisfies this would a piece wise function that just maps onto every number. My guess is that for every sequence of numbers I can atleast create a function that can correctly describe analytically at least half of the sequence.
Here comes the interesting part such an algorithm is basically the building blocks for an Ai. An example is text generation. Instead of using vast computational resources to adjust weights in a multidimensional NN. I can have a very long list of analytical functions that do the same thing reduce redundancies in the list and it would probably perform on the same level as current LLMs
Anonymous at Wed, 16 Oct 2024 18:25:29 UTC No. 16434867
>>16434100
Suppose that you have a sequence of i terms [math]n_{1}, n_{2}, n_{3}, ... n_{i}[/math]. The algorithm to create an analytic function for this is simple: multiply them all together to make a polynomial, because all polynomials are analytic.
Anonymous at Thu, 17 Oct 2024 02:42:22 UTC No. 16435642
midwit spotted
https://en.wikipedia.org/wiki/Polyn
Anonymous at Thu, 17 Oct 2024 09:20:21 UTC No. 16436003
>>16434867
That's not a minimal representation.
Anonymous at Thu, 17 Oct 2024 09:26:55 UTC No. 16436005
>>16436003
>>16434867
For example with a given finite fibonacci sequence the algorithm will either approximate the fibonacci polynomial formula or the recursion formula.
Anonymous at Thu, 17 Oct 2024 09:29:25 UTC No. 16436007
https://www.youtube.com/watch?v=iol
She knows
Anonymous at Fri, 18 Oct 2024 16:39:20 UTC No. 16438189
>>16434100
maybe.
Anonymous at Sat, 19 Oct 2024 17:27:08 UTC No. 16440002
>>16434100
you could say that again *chuckles*