To be specific they use one of the ViT-L/14 models. This type of labeling models have been around for a long time. They used to be called text-from-image or some other similar verbose description.
If the current generative models can produce porn then they can also produce CSAM, there's no need to go through another layer. The issue with models trained on actual illegal material is that then they could be reverse engineered to output the very same material that they have been trained with, in addition to very realistic generated ones. It's similar to how LLMs can be used to extract potentially private information they've been trained with.
>they claim in that screenshot to be "open source" Their setup is an elasticsearch database that they populate using undisclosed methods. Their "backend" (smat-be) is a wrapper that just forwards requests to their elasticsearch instance. You're supposed to contact them so they provide you with connection details in the form of an .env file that you then plug into the python wrapper "backend". Then they have a frontend that runs in vue which is a full blown web app and does most of the analysis. For instance the social graph to visualize connections is done client side in javascript. They get around the GPL by segmenting the data along with the collection process from the webapp part.
I think that it would cause immense butthurt if someone would use their code on a database populated scraped from shitty fedi outboxes. But it's too much work for just a little trolling.
>but your best bet would be looking what tunnels Chinese use. I looked a bit into this and found something that you may find interesting. It's a pretty good write up on how the Chinese firewall works, pretty recent too (April 2023). I found it very interesting that they ignore UDP traffic completely. https://gfw.report/publications/usenixsecurity23/en/
I can understand blocking popular VPN providers. But blocking VPN software, either openvpn or wireguard, will cause problems with firms using this kind of software for legitimate uses. Even more so if someone tries to block ssh.
I've looked a bit into shadowsocks and dnstt before and I'm sure they work much better than ssh tunnels. But it's really about obfuscating them, like the http obfuscation of v2ray, or GoQuiet shadowsocks addon, or routing everything through obfsproxy.
>blocking VPNs How does this work for business in Russia that use VPNs as part of their operation? Is the двач /s/ thread a good place to be informed about developments on centralized blocking in Russia? Speed considerations aside, can't people use an ssh tunnel? Surely they can't start blocking those without their own infrastructure falling apart.
I added basic emacs editing and movement shortcuts to mine. But the biggest problem is when you are trying to read something that is written badly. I sort of like the github code viewer too (but I feel dirty when I use it). Most people pay far too much attention on the editing part instead of the reading and understanding part.
That's horrible, these people deserve the worst. Are there any animal shelters/animal welfare organizations in your area, they might be able to help with the cat. As long as there aren't any problems with the organs a vet should be able to help.