Comment by Retr0id

3 hours ago

I've set up a socks5 "proxy multiplexer" that routes requests to different upstream proxies based on the request hostname. For example reddit routes via a VPS in Dublin, and imgur routes via Tor. I believe socks5 is the ideal layer to do the multiplexing at, for web traffic, because the request hostnames are visible to the multiplexer even if ECH/ESNI is in use. It was a oneshot vibecoded solution but it's been pretty solid thus far, so maybe I should open-source it.

I wrap the outbound sock5 traffic in mTLS, so it should look "normal" to anyone packet sniffing (not obvious proxy/VPN traffic), even though stealthiness isn't part of the threat model at the moment.

3 comments

Retr0id

Reply
bigiain  34 minutes ago

Perhaps consider putting it in public domain instead of using an "open source" license?

There's a decent legal ethical argument that LLM output isn't copyrightable, and for me a "one shot vibe code" definitely _isn't_ "your creative work", so the copyright that open source licenses rely on probably doesn't exist there.

I wonder if a new category of "non copyrighted shared source code" needs to exist for people who use Gan AUI to create genuinely useful software which would ne a net positive to society if shared, but that doesn't risk murkying the waters and undermining the copyright basis that licenses like GPL and Apache and BSD and MIT rely on?

  • farnsworthfusor  18 minutes ago

    If LLM output isn't copyrightable then it's already public domain, even if you say it isn't - if this is true you can just ignore the wishes of the person who thinks they're the copyright holder.