sounds great. those of us who work in industry understand thats not how any of this works and AI is the standard for development now. What the “AI Slop” People don’t understand is that development with AI depends on the person using it. You cannot give a chess novice AI and make them a master but if you give a chess master AI they become unstoppable. I wrote about this exact problem https://themildtake.com/articles/2026-04-18-ai-development-communication-problem/ here. I tend to mostly do the same thing as you I run a forgejo instance and mirror upstream to github as a backup.

well, it may be a hammer and nail thing for me (I work with compression a lot). But different compression algorithms are good at different things. Some solutions do better with text some solutions do better with structured data and some have special features that allow for enhanced compression with better tooling. GZip vs Brotli vs Zstd is a great example. Gzip is good at compressing small, structured data quickly where brotli accells at large text data and zstd is better at structured data and can be enhanced through training on that structure.

I’m pretty sure its repack they are trying to find the best way to make it as small as possible.

Well. ACTUALLY, it’s very simple you see every time you spin up a new linux box a little piece of Linus Torvold’s soul get’s imprinted in the CPU registers and then that fragment handles everything from there.

Because both major GPU manufacturers are controlled by the same family. The primary model repo is controlled by that same family now. Both major pipelines are controlled by that same family. ONNX and the diffusion pipelines are maintained by Microsoft. The alternatives are controlled by hyperscalers (with the exception of civitai I guess). All frontier model pipelines are directly controlled by a single family in China. So its very much possible that you could be banned from accessing them yes. We really don’t need a pirate instance we just need AN instance at this point.

I had a problem with DNS over TLS with PIA this sounds very similar. The conflict was DNS encryption type. I use Gluetun infront of PIA and the DNS resolver defaults to DNS-over-TLS on TCP 853. Through PIA’s tunnel, port 853 is blocked or doesn’t work reliably. SO I switched DNS to DNS-over-HTTPS on port 443 (I used Cloud9), which PIA allows and Gluetun can actually use. Not sure if this is you issue at all but if you have a common ARR stack with a vpn through gluetun this very possibly could be