← Home ← Back to /g/

Thread 106254354

24 posts 6 images /g/
Gnostic anon No.106254354 >>106254461
I wonder if somebody could script an AI-based network similar to Tor that doesn't require internet, it uses the AI as the server itself, so computers can interact without the need for network access because the AI is the network? It could be a browser plugin or a separate app, and installing it will either let you connect to the regular internet or a new network altogether and nobody will know you're online.

Think about it, there already exists AI protocols like Llama that can communicate offline by using the processing power stored on the device locally. But even though it's slow, there are ways you can make it faster. Write an open source script, route the traffic through custom generate proxy channels it assigns to a device (X25519 + ChaCha20 + Poly1305 + Kyber w/ no less than 261 bits of encryption, scripted in Python, using a similar customizable web experience to Hyphanet, but made user friendly, thanks to AI), make it a lightweight application to reduce processing power used, make it distributed like chatgpt where eacj computer serves as a node without piggybacking off preexisting corporate infrastructure, and the app itself is the ISP. Small, distilled versions of Llama and MPT can already run locally without relying on GPUs, and it could even be used as a backdoor to the regular internet.

This would dethrone not just Tor, but any ISP out there along with that. This is every NSA and tyrant's worst nightmare some shit like this. You don't even download it online or rely on any corporate-owned infrastructure. You could just script your own parasitic network, powered by AI.
Anonymous No.106254373 >>106254418 >>106254459
an actually smart idea on /g/? holy shit
Anonymous No.106254381
I'm having GPT-5-Thinking-High draft up a whitepaper for the idea right now. I'll let you know when it's done.
Anonymous No.106254418 >>106254444 >>106254473
>>106254373
No, this is the dumbest thing I've read here in a while what are you talking about?
Anonymous No.106254444
>>106254418
you just don't get it
Gnostic anon No.106254459 >>106254473 >>106254732
>>106254373
The AI bot itself accesses the internet for you and pulls results online. Your computer is the node itself and it's an AI-based meshnet protocol that doesn't piggyback off of preexisting infrastructure. It's a parasitic server that fetches data and connects you through backdoors.
Anonymous No.106254461 >>106254519
>>106254354 (OP)
>so computers can interact without the need for network access because the AI is the network?

lol
Gnostic anon No.106254473
>>106254418
See >>106254459

I've studied computer science, cybersecurity, some programming, and chow computer networks run (+some grayhat stuff) and I can tell you you are wrong. Llama already works offline. It acts as a separate machine, that uses your device to provide results. It's an open source AI script. You can look it up online.
Gnostic anon No.106254519 >>106254527
>>106254461
That's not what I'm saying. The AI is the ISP. All an AI does is detch resiltz from the preexisting web for you, so it could serve as an alternative to a cloud server.
Gnostic anon No.106254527
>>106254519
*fetch results

The NSA wouldn't even know you're online either.
Anonymous No.106254534 >>106254599 >>106254744
You're the same kind of retarded person thinking they could use the "blockchain" for everything without actually having a fucking clue how technology works.
Gnostic anon No.106254599 >>106254630
>>106254534
Wait, actually the blockchain might add a nice touch of decentralization to the AI "network".

Good thought, even if meant in jest.
Anonymous No.106254630 >>106254744
>>106254599
You're fucking stupid
Anonymous No.106254679
I am a scientist and I think this is brilliant.
Anonymous No.106254732 >>106254744
>>106254459
that's not how it works. the llm just runs commands based on your input that fetches shit off the internet just like how a browser, or any other network software, does.
Gnostic anon No.106254744 >>106254770 >>106254796
>>106254534
>>106254732
>>106254630
The AI is still needed to give instructions to the API and you can make it a fully distributed lightweight AI software everytime there's a pull request — the API doesn't know what to do by itself, it's just human + AI + API = offline server access.
Gnostic anon No.106254770 >>106254796 >>106254819
>>106254744
So when you type in "www.youtube.com" the AI tells the server that there's a pull request, so the API will simply fetch those results back to you in real-time. There might be a delay, but if you run this as a lightweight client, it won't require a quad core desktop PC processor with four fucking 200 TB RAM sticks just for a page to load in 15 minutes. Or the AI could simulate another machine becaise your computer is not actually connected to the internet, so all they'll see is bot traffic as a subnet mask.
Gnostic anon No.106254796
>>106254770
>>106254744
Bingo. I made some mistakes in the OP.
Gnostic anon No.106254819 >>106254910
>>106254770
It could use SOCKS5 proxies to spoof approved ones in a network without having to rely on DTN and manual access. But that's just in theory. I don't recommend this. Thd only legal way of doing it involves setting up your own infrastructire (so they can still spy on you). Hence why you still need proxies to gain resources for your computer to even access the WWW.
Gnostic anon No.106254910 >>106254943
>>106254819
Or you could mirror a server, more preferably. That would satisfy the privacy aspect and the legal issues since it's no longer technically a backdoor. Let's say Bulletproof VPS services let you mirror their server without being physically connected to it. Then it would no longer be backdoor access. It would cache static pages and the only real issue then would be dynamic (like livestreaming). In that case certain asspects of the web wouldn't work, but you can still get online via server cache.
Gnostic anon No.106254943 >>106254953
>>106254910
Assuming you're emulating a server and you already have that server's permission, any kind of traffic routed through that server would already be available w/o a manual handshake process I'm assuming, but as for how you would emulate outbound traffic or how your computer would appear, I'm guessing you would simply drop, I doubt that you would even need to emulate traffic also unless it were for niche purposes. The only problem remains dynamic content (and that includes outbound traffic unless it were contained within that network and didn't communicate out).
Gnostic anon No.106254953 >>106255034
>>106254943
I mean it's possible to communicate out, but there's no legal ways of doing it that are practical.
Gnostic anon No.106255034 >>106255085
>>106254953
The only real bypass this is to start over from scratch because the internet is too centralized. Ever since the 1996 CDA and DOTCOM explosion after the internet started to become popular in the mid-1990s, the government has stuck its claim over the WWW and turned it into an Orwellian surveillance platform for the government to see every last thought that goes on in a person's mind. Before that, websites didn't have ".com" suffixes, and a link would appear as "http://www.greenplants//" instead of "http://www.greenplants.com". Back in the day, people didn't have to worry about fake sites, bitcoin miners or Google Voice recording you strain on the toilet in the privacy of your own home. Nowadays you have to go through some intense hurdles just to maintain any privacy on the web. Back then "cloud computing" didn't exist. Making a website was free. The first web browser was released in 1990 called WorldWideWeb, and it was an interface that even let you edit/vandalize people's webpages. DDOSes weren't a problem then either, because of bandwidth issues. I don't even know how piracy worked back then, but sometimes people were able to pirate video games, and that was considered a big deal. Back in the 1970s, emails took as ling as snail mail did to reach somebody, and oftentimes they would get "lost" or never send. The modern internet requires resources for you to even connect otherwise it's a "cybercrime". The internet at firzt didn't even use TCP/IP, it used some outdated protocol, and even in the early days of IP addresses, your OS would refresh them for you - they weren't used as identifiers, just entry points for a user. American privacy laws are shit. In other countries like Germany or Canada, IP addresses enjoy the same legal protections as a SSN. There needs to be an altnet that can satisfy both the flair that the old internet had, but modernize it.
Gnostic anon No.106255085
>>106255034
The closest thing to an "offline net" that I know of are networks using protocol stacks above IPv6, which ISPs don't even support because they're blacklisted and the only way you could connect to them is if you had custom firmware on your router so you could force the connection, but that's about it as far as I know. While GNUnet by itself isn't one of these networks, however it could let you interface with them. But that would imply creating a parallel internet, which is doable in theory, but it's typically used by hackers to do shit you wouldn't think was possible in networking.