🧵 Logarithm algorithm
M'ens. Lem Daier at Fri, 19 Jul 2024 03:15:50 UTC No. 16288017
Hi
Suppose you could take the logarithm of an entire file. First, how do you do it? And second, what are the benefits? Is this a project I should embark on? The pay is trash but what about recognition? Why would a PI even create a project like this? He won't even tell me. That's why I'm scratching my head. Promises huge fame.
Yoda Biden at Fri, 19 Jul 2024 03:27:37 UTC No. 16288023
>>16288017
For example person will reply to the same his whole life
M'ens. Lem Daier at Fri, 19 Jul 2024 03:30:43 UTC No. 16288025
M'ens. Lem Daier at Fri, 19 Jul 2024 05:45:38 UTC No. 16288109
>>16288017
OK I realize it now.
Addition of log is multiplication, subtraction is division. So O(1). I.e.: send huge files as a polynomial of sorts.
Anonymous at Fri, 19 Jul 2024 08:22:11 UTC No. 16288196
>>16288017
I'm confused, take the logarithm in base 2 of the binary representation of a file? Just find the position of the first 1, and the total # of binary digits succeeding it and you have an easy lower bound. What level of precision would you need it to be, how fast does it need to be? Why would such a simple problem give you "huge fame"?
M'ens. Lem Daier at Fri, 19 Jul 2024 08:44:43 UTC No. 16288204
>>16288196
>>16288196
>Why would such a simple problem give you "huge fame"?
Exactly my point.
Why is this such an easy problem, and yet unsolved? How do you calculate the logarithm of an entire file to begin with? And yes, it needs to be a precision that's good enough so that I can send you a short number and you get a video file to watch. Let's say: 2^35 / (5^7+1) or something.
And then you get a big file. But supposed to be an MP4 file or something.
How come it's such an easy task, and yet: it doesn't exist? Do mathematicians not know of computers? Makes 0 fucking sense mate.
So really: we're "just" extracting the prime numbers.
Anonymous at Fri, 19 Jul 2024 08:50:28 UTC No. 16288207
>>16288204
>, it needs to be a precision that's good enough so that I can send you a short number and you get a video file to watch
That is mathematically proven to not exist. If you're interested in videos specifically, there are tons of nonlossy compression algorithms (and lossy algorithms 99% of people won't notice)
M'ens. Lem Daier at Fri, 19 Jul 2024 08:58:18 UTC No. 16288214
>>16288207
That's not how it works my dude.
You send ME a key/number K. I send YOU a polynomial P. I just assumed an entire file to be 2^35 / (...) and that's it. I didn't say it was MP4. I didn't say it had images of any kind.
These are two different subjects. Here, we're talking about YOU send me YOUR KEY.
And I send you a file P in the format of
22 x K^23 + 153x K^21 + ... K^20 + K^1
It makes universal languages absolutely possible. Without it, ChatGPT doesn't exist. Every meaning or reason is relative. And this is a proof you cannot deny. Because I can divide a file by 3 and it'll be 2 bits shorter. Maybe you'll have a carry. But that's the reason why we want to factor the entire thing. And that's where logarithms come in handy. We need to figure out what fraction this is.
Anonymous at Fri, 19 Jul 2024 09:19:25 UTC No. 16288224
>>16288214
>You send ME a key/number K. I send YOU a polynomial P
Are you talking about file hashing? That's achievable, but you need to store the entire piece of data somewhere
>It makes universal languages absolutely possible
What universal languages? Is this some kind of information theory jargon I've never heard of
>Without it, ChatGPT doesn't exist
What? ChatGPT is just a bunch of matrix operations, it doesn't need some kind of universal compression algorithm to work
>I can divide a file by 3 and it'll be 2 bits shorter. Maybe you'll have a carry
The issue is that you have to record each of the steps to recreate the original file even if you have the shortened version. That winds up being the same length as the original file in the worst case (unless you make certain assumptions about the original files properties)
M'ens. Lem Daier at Fri, 19 Jul 2024 10:15:49 UTC No. 16288250
>>16288224
It's not file hashing. It's about studying numbers. PIxels are numbers. That means we can index them.
Concatenate latitude, longitude, unix timestamp, image number, x, y as your input variables and the pixel value as 32 bit is your output.
Clearly, you will have many pixels that are the same? So you can do conjunctive normal form to create logic, to reduce terms.
That makes sense, right? By that, I mean it rings a bell or resonates inside you. It has a meaning in relation to something you know already, so you can derive what I mean using 2 or 3 words.
I tell you: cinema, safehouse
Probably our regular theater. But when? So that I see if there's a bell that rings when we set the time. But it already makes sense what I want to tell you: I want to go to cinema with you to watch Safehouse. I don't need an entire sentence.
Likewise, I don't need an entire set of pixels. I just need to know where linearity stops. And when linearity stops, can I derive a binary logic equation?
Do you see where I'm going with this?
Anonymous at Fri, 19 Jul 2024 11:31:50 UTC No. 16288277
>>16288250
he wants to describe the whole phase space of a computer (processor inputs, outputs, ram, rom, motherboard, hard drive and video outputs) as a binary semi-Thue grammar you can do boolean algebra and optimization on. somehow he thinks this is new shit? you can't do conjunctive logic to join "equal" pixels cause they aren't equal, you can't compress pixel index/coordinate information, just represent it in other ways (binary, for example).
>concatenate all video metadata and compress it
tell me, how is this new?
>cinema, safehouse
you're using non-monotonic speech to share information. there's unwritten context underlying the symbols.
M'ens. Lem Daier at Fri, 19 Jul 2024 12:22:44 UTC No. 16288320
>>16288277
I want a black box I input metadata in and it gives me pixel data back.
You people can't seriously be THIS dense. This is literally the averaging of the two nearest neighbours. The third variable tells you where between the two you are.
96 = (128+64)/2
So 96 is exactly between 64 and 128. I tell you: I'm thinking of a number that's between 64 and 128. By default, you assume it's exactly in the middle. I.e.: 96. But what if I told you it"s 1/4 of the way. You have more information. Your data is more precise.
Kinda like: seen one suburb, seen em all. So someone can tell you a story about where they grew up and you can immediately picture a mental image of what they are telling you.
I.e.: you have a reference point. The story is relative. The house and the suburbs and everything that makes sense to you is in relation to something that originally made sense to you. Perhaps your name and that's why I think people are stuck on identity.
It's hard to describe if you don't let your guard down enough to stop ridiculing the idea subconsciously and instead just dive in for a moment.
Just read what I wrote above carefully and literally without any added context. I try to write in a way that makes AI remember the entire context at all times. However: clearly, if we used our human brain power to figure out 1st, 2nd, etc order kinetics of waves, I'm pretty sure we could reconstruct the universal time independent wavefunction. I.e.: see the past from waves that are still residual from being stretched by the spin of the Earth.
I sound incoherent if your guard is up. But if you close your eyes in between sentences, you will understand what I mean exactly. It just makes so much sense. We have free will. And we have probabilistic wavefunctions at angstrom scale. What is missing is a macro view is all. Generalization. Unity and symmetry instead of distinction and separation.
Anonymous at Fri, 19 Jul 2024 14:16:24 UTC No. 16288409
>>16288320
>we have free will
we have no meds, clearly
you're subconsciously reinventing the wheel of image compression while not actually adding anything new, thoughbeit
Anonymous at Fri, 19 Jul 2024 20:53:24 UTC No. 16288837
>>16288320
>The third variable tells you where between the two you are.
>96 = (128+64)/2
>So 96 is exactly between 64 and 128. I tell you: I'm thinking of a number that's between 64 and 128. By default, you assume it's exactly in the middle. I.e.: 96. But what if I told you it"s 1/4 of the way. You have more information. Your data is more precise.
>Kinda like: seen one suburb, seen em all. So someone can tell you a story about where they grew up and you can immediately picture a mental image of what they are telling you.
This is basically ai and interpolation.
Between two coordinates in the real plane, dont matter how close, you have a non numerable infinite number of curves that join both points.
Interpolations always lose information of the original, you can only approximate the real value.
M'ens. Lem Daier at Sat, 20 Jul 2024 03:57:24 UTC No. 16289214
>>16288837
Exactly, interpolation is the word I was looking for. Unfortunately, it's difficult for me to express myself, so sometimes it's easier with 2 or 3 words.
And yes, you lose some. You also lose to DCT or whatever. I'm not looking for image comrpession. I'm looking for actual sites of magnetic radiation, absorption, re-emission, etc. to predict stock market signals.