7 ms·

This take is borderline absurd. The risk of AI is in centralization of AI. In the same way the risk of the web was in centralization of the web. Those very re

This take is borderline absurd. The risk of AI is in centralization of AI. In the same way the risk of the web was in centralization of the web. Those very real risks manifested in very real ways on the web. I have young kids. I'm not exaggerating when I say I wake up every day and diligently build the future I was promised for them. I've got somewhere around 10 years, conservatively, until Instagram/TikTok/YouTube/Facebook/etc. try to get my daughter to kill herself. I've got around 10 years until she is a statistic, hopefully the statistic that survives. Social Media wasn't always Social Media. We weren't promised Social Media. We were promised Social Networking. Social Networks were supposed to connect us with our communities. Bring the world together in a way previously impossible. Allow the free exchange of information and ideas. That future is still possible. But it turns out expecting some dude in Palo Alto to pay to be the free Proxy and Archivist of the entire world's social communication is a hard problem to solve. It's a hard engineering problem. It's a hard financial problem. It's a hard legal problem. It's a hard ethics problem. Social Networking companies didn't solve that problem. They failed and pivoted into Social Media. Social Media divides us. Social Media isolates us. Social Media believes my friends, my family, and my community "aren't engaging enough" and litters my timeline with randos it's trying to hype to keep me engaged so they can monetize the experience. I go to my PTO meetings and parents talk about their kids being isolated and depressed. They talk about body image issues and lack of purpose. I go to my community gatherings, business networking groups, and gym. People talk about how lonely they feel. How angry they feel. How depressed and anxious they feel. Instead of the Social Networks we were promised, we have these Social Media monstrosities that take randos on the internet and try to make them famous in front of other randos on the internet in a variable rate reward slot machine. Social Networks on P2P technology have been viable for at least 5 years now, but the industry hasn't caught up. It's possible to join a p2p social network from your browser tab, pair it with your phone, have them both simultaneously be able to generate posts and sync over a gossip protocol without conflict. And it's possible for your social group to backup your content. All of this is still possible with the Application Service Provider architecture of having a "backup pinning server" that will make your data highly available. Identity is solvable in a way that allows for full control of your identity without relying exclusively on public/private key cryptography - and permits social recovery of your identity when your keys are lost or compromised [1]. End users don't need to understand private keys or public keys. They just know they have an account, and that their friends can send them a friend request. I'm going long on p2p applications. I think they aren't just viable, but that they can out compete the user experience that current generation Application Service Providers offer while simultaneously breaking the incentives that are destroying the societies my children will inherit. AI is no different. The call to arms about AI regulation and alignment often degrades to advocating for a world where my children do not get access to AI because it's "too dangerous" for them - while someone else's kids belong in the privileged AI class. The trajectory forming for AI strongly signals to me that it will be the single most profound means of production our species has produced. Any philosophy that advocates taking that away from my kids can go play in its own corner. I'm not taking back the web for my kids just to let them take away AI in return. If these things interest you, and you're interested in doing hardcore research on productionizing real, no bullshit, peer-to-peer networks both for end users and enterprises, my email is in my bio. I want to talk to you. We need more good people in this space, and we have budget for it. My team is building peer-to-peer networks for AI, but the solutions that break AI out of it's ivory walls aren't AI specific solutions. These peer-to-peer building blocks are general purpose and the next generation of the web is going to run on them. This is the other other road ahead [2]. We will succeed or I will die trying. [1] http://www.blankenship.io/essays/2023-09-24/ http://www.blankenship.io/essays/2023-09-24/ [2] http://paulgraham.com/road.html http://paulgraham.com/road.html

cmrdporcupine 3y ago
Everything you're saying resonates, though I'm in that "10 years" zone, ahead of you, in regards to parenting. I'm here to talk about the problems of adolescent girls in this era, if you ever want to reach out. A horrible reality I've lived as a parent. And I would like to build the online social world we wanted, back in the early 90s, rather than the one we eventually sold ourselves to, or were sold to. ... and to me that social tech we need is concrete rather than virtual. It's immanent rather than transcendent. Infra or "within", rather than meta and "without", or virtual. Its creative rather than passive-consumptive. And narrative-textual-descriptive rather than artificial-graphical. And to me the outline of that looks something like what was being grasped at in the early 90s with synchronous, live, shared creative social worlds like LambdaMOO -- if that isn't too obscure for you -- than what we ended up with the web, and then Facebook, Twitter, etc. The peer to peer aspect is of less ... interest? ... to me? I'm not convinced that this aspect of it is as important as it is to refocus on control/ownership, intimacy, and moderation?
- r3trohack3r 3y ago
  > Everything you're saying resonates, though I'm in that "10 years" zone, ahead of you, in regards to parenting. > I'm here to talk about the problems of adolescent girls in this era, if you ever want to reach out. A horrible reality I've lived as a parent. This is heartbreaking and I'd love to talk... Permission to reach out via email? > I'm not convinced that [peer to peer] is as important as it is to refocus on control/ownership, intimacy, and moderation? I believe these are synonymous. > control/ownership When you invert control/ownership, the user controls their data. The application is delegated access to it. This is a peer-to-peer data structure. Instead of generating the data on an ASP's server, the user generates their data and stores it in a data structure they have control over. They can _delegate_ generation of data to an ASP style app, but ultimately the app is taking actions _on their behalf_ which the user owns. They can pull that data down to their laptop, push it up to a peer-to-peer enabled NAS (purchased from the big box store) that they plugged into their router, etc. The app can't revoke access to their data if they've backed it up. The app can refuse to continue mirroring it or modifying it, but it can't _lock them out_. It's the user's data. That data is sharable and discoverable without the app's permission. The app can participate in sharing and discovering the data, but it can't _block_ the user from interacting with other users on the web. The user isn't _forced_ to interact with other users either. They are in full control of who they associate with. Right now we are solving this with a hash-tree. Every time a user takes an action ("creates a post", "uploads a picture", etc.) it's added to their hash tree. When you add a new device, the tree "branches" and you get two branches, one for each key to sign and add next blocks. This solves concurrent writes much like a CRDT. That hash-tree is an append-only database. Anything you want to record goes into the database. The nodes in the tree do not store content, only a _reference_ to the content. This keeps the tree very small (1M nodes can fit in less than a GB of storage). You can backup the entire tree without backing up all of the content. And you can "delete" large files without having to rewrite the tree. The tree itself isn't the user's identity though. That's fiat. If a key is compromised, users can flag content and say "hey, this isn't Alice, Alice wouldn't send me a crypto scam." Users can take two trees and "merge" them under the same identity (these are both Alice, but Alice lost her login and got locked out of her account, so has a new tree). This can be done without appealing to some application owner to "permit" you to fix Alice's social identity from your perspective on the web. Then we use a capability system to delegate rights to keys added to the tree. So you can add an app (like iCloud) to your tree, and that app can now take actions within the scope of the capability it was granted. > intimacy, and moderation Intimacy, to me, means letting users decide how they want to model their social interactions. Who they pull in content from, how that content gets aggregated and displayed, etc. Moderation, to me, means deciding what the user gets to see. Those decisions are for communities, not for some remote disinterested platform operator. You solve intimacy and moderation by allowing communities to form and moderate themselves.

cmrdporcupine 3y ago

Everything you're saying resonates, though I'm in that "10 years" zone, ahead of you, in regards to parenting. I'm here to talk about the problems of adolescent girls in this era, if you ever want to reach out. A horrible reality I've lived as a parent. And I would like to build the online social world we wanted, back in the early 90s, rather than the one we eventually sold ourselves to, or were sold to. ... and to me that social tech we need is concrete rather than virtual. It's immanent rather than transcendent. Infra or "within", rather than meta and "without", or virtual. Its creative rather than passive-consumptive. And narrative-textual-descriptive rather than artificial-graphical. And to me the outline of that looks something like what was being grasped at in the early 90s with synchronous, live, shared creative social worlds like LambdaMOO -- if that isn't too obscure for you -- than what we ended up with the web, and then Facebook, Twitter, etc. The peer to peer aspect is of less ... interest? ... to me? I'm not convinced that this aspect of it is as important as it is to refocus on control/ownership, intimacy, and moderation?

r3trohack3r 3y ago
> Everything you're saying resonates, though I'm in that "10 years" zone, ahead of you, in regards to parenting. > I'm here to talk about the problems of adolescent girls in this era, if you ever want to reach out. A horrible reality I've lived as a parent. This is heartbreaking and I'd love to talk... Permission to reach out via email? > I'm not convinced that [peer to peer] is as important as it is to refocus on control/ownership, intimacy, and moderation? I believe these are synonymous. > control/ownership When you invert control/ownership, the user controls their data. The application is delegated access to it. This is a peer-to-peer data structure. Instead of generating the data on an ASP's server, the user generates their data and stores it in a data structure they have control over. They can _delegate_ generation of data to an ASP style app, but ultimately the app is taking actions _on their behalf_ which the user owns. They can pull that data down to their laptop, push it up to a peer-to-peer enabled NAS (purchased from the big box store) that they plugged into their router, etc. The app can't revoke access to their data if they've backed it up. The app can refuse to continue mirroring it or modifying it, but it can't _lock them out_. It's the user's data. That data is sharable and discoverable without the app's permission. The app can participate in sharing and discovering the data, but it can't _block_ the user from interacting with other users on the web. The user isn't _forced_ to interact with other users either. They are in full control of who they associate with. Right now we are solving this with a hash-tree. Every time a user takes an action ("creates a post", "uploads a picture", etc.) it's added to their hash tree. When you add a new device, the tree "branches" and you get two branches, one for each key to sign and add next blocks. This solves concurrent writes much like a CRDT. That hash-tree is an append-only database. Anything you want to record goes into the database. The nodes in the tree do not store content, only a _reference_ to the content. This keeps the tree very small (1M nodes can fit in less than a GB of storage). You can backup the entire tree without backing up all of the content. And you can "delete" large files without having to rewrite the tree. The tree itself isn't the user's identity though. That's fiat. If a key is compromised, users can flag content and say "hey, this isn't Alice, Alice wouldn't send me a crypto scam." Users can take two trees and "merge" them under the same identity (these are both Alice, but Alice lost her login and got locked out of her account, so has a new tree). This can be done without appealing to some application owner to "permit" you to fix Alice's social identity from your perspective on the web. Then we use a capability system to delegate rights to keys added to the tree. So you can add an app (like iCloud) to your tree, and that app can now take actions within the scope of the capability it was granted. > intimacy, and moderation Intimacy, to me, means letting users decide how they want to model their social interactions. Who they pull in content from, how that content gets aggregated and displayed, etc. Moderation, to me, means deciding what the user gets to see. Those decisions are for communities, not for some remote disinterested platform operator. You solve intimacy and moderation by allowing communities to form and moderate themselves.