I have avoided AI (LLMS) because it didn’t seem nearly useful enough to warrant supporting these tech assholes. Is there anyway to use it locally on a normal device (16 Gb ram, 8 year old Geforce graphics, etc?). Can I just slow it down and get it to work locally?
Starve the beast by VPN hopping on opencode free models. It’s become a little more difficult in recent weeks because they are now identifying people doing this via MAC address, but it’s still fully possible.
Your MAC address is not transmitted on TCP/IP packets, it’s only on layer 2 of the OSI model. Also, you can change your MAC address at will on most OSes, but again, it’s only used on your local network and cannot be seen once your packet traverses an IP router. They’re cataloging VPN exit nodes, if anything.
Mac address isn’t needed anyway. Fingerprinting is very sophisticated now. Obfuscation of your host is nearly impossible. Many technologies exist that can look at maybe five things to fingerprint you. Things you can’t change like timing of various events; a few little probe JavaScript object to see how quickly your machine renders an object will be very unique to you.
Oh, it’s absolutely possible to obfuscate your fingerprint.
You know you’ve done it when you get banned from websites immediately after sign up. Websites that sell your account data expect to uniquely identify you. And if they can’t, immediate ban.
No, it’s not transmitted on TCP/IP packets, but the way opencode works makes it so it can be part of the pinning data. Along with probably a hashed version of the context attached to the project and other things. I have read a bit about it and vibe coded some workarounds and it’s not as simple as them cataloging the VPN exit nodes, because then none of them would ever work. I have like a “scramble” command on my setup now to get around it. This was all implemented by them sometime in the last month.
Before I added the scrambling switching between VPN nodes after the october limit pinning would immediately associate you with the old limit when switching between VPN nodes, and after scrambling the MAC address and other things (nuking project metadata) it picks up a new limit again. I’m not 100% sure what specific pieces of metadata they’re using for pinning because I decided to just have the sandbox code scramble them all at the same time, but they are pinning now.
The easier fix would be for them to just require an account, and I suspect eventually they’ll do that (I suspect they like the anonymous for demos, but that’ll only last so long). I also suspect that google AI mode will eventually get the same treatment (account requirements and usage limits) because I find it very difficult to believe someone hasn’t already Jerry-rigged up a coding harness attached to Google’s free AI mode.
Simply using ChatGPT for free is starving the beast. Couple of months ago Altman said he was surprised that the new $200 tier was still losing money.
So I’m fine with free LLMs. Every use costs them and brings the bubble bust closer.
I find that hard to believe honestly. How much of that is marketing to make it seem like greater value-for-money to customers deciding what tier to use?