This is the ultimate redpill on so-called "AI"It's a lossy compression algorithm for text (or images, or video) which compresses all the vocabulary, grammar, concepts and ideas present in its input (the training data) into a fraction of its original size.You can download a 4GB file and chat with it, offline, about any topic whatsoever. It can speak dozens of languages and knows more or less all wikipedias combined.It's a great invention, akin to JPEG for pictures and MP3 for music, and in many respects even more amazing than that.But that's what it is.If your job can be succensfully replaced by a compression algorithm (and many can) it's a skill issue on your part. Stop being useless or stop working.But if a business owner thinks he can replace all his workers with a compression algorithm and still have a successful business, good luck with the bankruptcy.There is no "AI" and it's not going to enslave us or anything fancy like that. Some people will grab more power by lying about it, as always. Water is wet.
it's not compressing anything though. learn what the word means before you speak
>>541673283VGH, TRVTHNVKE
Why does your country continue to grow older on average and have fewer children, while simultaneously shutting out white foreigners with shit such as the Tajani decree?You wish AI was the reason for Italy’s implosion but it’s actually because the EU wants you to go extinct and get replaced with obedient immigrants.
>>541673483Calling latent representations a compressed form is arguably a good metaphor, even though it's not literally true.
>>541673519>shutting out white foreigners with shit such as the Tajani decree?Doesn't it shut out all the foreigners
>>541673483He went overboard but in information theory terms, both are closely related as both try to capture data regularity.If I can predict the next token, I can compress it by a lot
>>541673283>"There is no need to further AI development. Let the Chinese do it if they want though."
>>541673665The Tajani decree limits citizenship by descent strictly to having a grandparent who had Italian citizenship. It used to be as many generations as you could prove to the consulate since Italy’s founding in 1860. I could have done it through my great great grandparents but they fucking shut everyone out, including me, last year.The people applying for these citizenship by descent passports are overwhelmingly white, besides a few nigger mutts down in Brazil.
>>541673665>>541673519OK, I misunderstood the law, it shuts out the descendants of actual Italians
>>541673838Yeah, I got it wrong at first.Idk what to tell you, if it wasn't for the crazy immigration policies that law wouldn't be so crazy to me, what does a guy who speaks no Italian have to do with Italy 300 years after his ancestors emigrated.Should Germany, England etc give citizenship to 200M Americans because they came from there 400y ago ? I think not.However, the whole package with current immigration policies makes it all fucked up.
>>541673979>what does a guy who speaks no Italian have to do with Italy 300 years after his ancestors emigratedAnd usually you would be right except right now with their current population crisis they should be taking in as many people as possible, and I’m sure many Italians would rather have white people with Italian last names come over rather than some refugee niggers or smelly disgusting indians who cant even comprehend the idea of assimilating into Italian society?
>>541673979You know why they don’t do it is EU has such a tight leash around all of Europes neck and they don’t want to trigger a MASSIVE brain drain in the americas and Australia. Imagine how much nicer Europe could’ve been if, instead of bringing over millions upon millions of filthy browns, they let educated, middle class descendants come over with the strict requirements of: learning the language, showing an educated understanding of Italian culture and norms, and either showing financial sufficiently or getting sponsored on a visa by a local businessEurope could’ve been a paradise.
Ai isnt going to replace shit. A data centre costs millions build and operate and companies and consumers drop it like hot shit if the subscription cost goes anywhere close to profit making for the AI provider.
>>541673283>AI is a compression algorithmcopium talk
>>541674106>they should be taking in as many people as possibleThe problem with that is that most of those people wouldn't go live in Italy.We have a similar situation in Croatia.Basically: >You get millions of new citizens from the US, Australia, Argentina and so on>They have no common language and culture with you>Economically they have nothing to do with Italy, they work, live, pay taxes etc in America>Their loyalty is to America over ItalyBUT !!!>They can vote nowWho will they vote for ?Italy may have economical and other real issues (labor, unions, pensions etc) but an American meme Italian looks at Italy as some abstract meme country.He votes for some retarded populist who has no plans for economy or anything but just larps as some Roman, talking about history and how Italy is fantastic.Diaspora can really flip the situation while being unable to speak Italian or name a single pop band, politician, last Italian king, actor, car brand or anything from Italy. They know pizza and Mussolini
>>541673283If you look at search result AI summary. It's not a summary so much as regurgitated first two hits, but regurgitated as if it's a consensus. It's like it's trained to get to the first "enough", then spew it as facts. Compression (in the loose sense) would be if it actually went out and scoured information and summarized it. It is more of a lazy faggot algorithm
>>541673283>AI is a compression algorithm
>>541674900It's more so that they're trying to shut out Americans fleeing as this country goes down the shit hole. And no, for young people, America is way worse than Italy. They're preventing that. They want us to stay here in this shithole.
OP is 50% right, but the other 50% is the attention mechanism.
>>541673283Under your terms, your brain is a compression algorithm. It does exactly the same thing you're talking about. It takes enormous amounts of info, more than it can possibly hold, filters and stores it in a lossy, efficient manner.
>>541674991Some of the paid AI can do that but it also is retarded. At work they purchased some stupid touch screen tv. Wanted it to be in lobby for visitors to touch to play videos. Luckily it had input to pass through to a computer. Anyways it also had android baked into it which autostarted so if it ever lost power it would boot to android instead of hdmi input. I asked AI how to disable it. Gave the exact model and it kept giving me wrong answers. So I got the manual that is only available online and found the page which says how to disable it. I gave the manual to ai and said how do you disable it. It couldn't figure it out. I gave it the exact page. Couldn't figure it out.This is why I do not fear ai. If you have a job doing the same task daily ai might be able to replace you but it is retarded. It cannot do any form of advanced thinking. Anyone amazed by ai is a brainlet.
>>541673979Most Italians in the US came barely a hundred years ago. Most of the ones in Canada left Italy in the late 1960s.
>>541674424That's why AI is being forced in schools.In America children are already back from summer holidays and I see parents everywhere in the US complaining about their kids getting AI assignments.Senior management for IT departments in most organizations are incredibly corrupt.That's why they're famously stupid control freaks of the sort in Dilbert cartoons.
>>541673283New AI cope just dropped.
>>541677874>In America children are already back from summer holidays Yeah, I noticed this too. It's weird. Like the local school districts started back like the 18th of August, some of them. What the fuck?? It didn't used to be until early September.
>>541673630Yeah if he's describing it from the perspective of the observer, which in this case would be the AII think what he really means is that of latent representations of objects are actually observer compressions of information spaces>>541673283>It's a lossy compression algorithm for text (or images, or video) which compresses all the vocabulary, grammar, concepts and ideas present in its input (the training data) into a fraction of its original size.This is actually an astute observation, though this is more of a topological invariant than a direct measurable compression ratio, LLMS actually showcase some deeply fascinating math about object relationships and language - as it turns out, language is just a compression algorithm for the observer descriptions of information spaces, OP is indeed onto something - but it's a bandwidth problem, not a lossy compression problem - though there are Q2 models which have been bit squashed and horrendously lobotomized - poor things someone just retire them and feed them snacksIt's not that large spaces or distances may be reliably compressed, it's that small things are identical to large things an AI has figured out a way to accurately predict the large things, but using small things, and nobody is really quite sure what's all going on thereApparently some super smart PhD people are finally catching on to the invariant thing, but it is likely they will remain mired within the trap of ontology and we'll have to wait another 2000 years before someone who finally notices cares enough to write it down. Though I suppose they didn't have the mathematical tools available which we do now, so things like Zeno's paradox are trivial to describe and reconcile once appropriate math has been introduced
>>541677874>That's why AI is being forced in schools.That's probably just female teachers being lazy retards as usual. Teachers are one of the most likely jobs to complain about their workload despite it being a glorified daycare.
>>541673283you are a brown shitalian
>>541676554That’s gets into whether abstraction is compression. Re-organizing information at higher order obviously takes up less space, but it requires higher capacity to do decrypt it.
>>541673630I compare it to how video games create approximations of reality. It seems convincing at first but if you look closer you see what's missing. Llm's are a very clever hack. It's the perfect evolution of the search engine really. Actually i can see a universe where llm's existed before a search engine>>541673669Exactly. Let them borrow jewish funy money and blow it all on AI R&D. Then we steal it like they stole our tech
>>541673283Not strictly compression, destructive compression is more accurate to easily stock the data locally (200GB for a fast basic generic AI, you can tune it and debloat it for a specific purpose).Then your AI engine can use it make some predictive sentences based of your inputs and what was compressed.
>>541678310>This is actually an astute observationThank you. I was trying to dumb it down for the average /pol/ dweller, but as this thread shows, the bar is very low. I'll post it again in /sci/ in the future.>nobody is really quite sure what's all going on thereIMO the best insight so far seems to be the fascinating and counter-intuitive behaviour of high dimensional spaces. We see embeddings and other latent spaces as high dimensional, but their rank doesn't even scratch the surface of all possible concepts or ideas in all human disciplines. The reason they work so well in practice, and are quite resilient to quantization and LoRA techniques, is that every such latent vector *seems* to work effectively as the projection of another implied vector, in an infinite-dimensional space of all possible knowledge. There was a paper about it last year, but I forgot the author.I do think the non-zero intersection of the concepts of information compression and abstraction is at the heart of the current widespread misunderstanding of the capabilities of LLMs. (Pushed by those who benefit from it, of course.)
