Sounds like you are talking past each other. GP is saying the harness of codex is higher quality, which I can believe, even if the models are not as good as Opus/Fable.
I'm on iPhone, i found a workaround involving having headscale set the dns using 'magicdns' or something and having that pihole join the tailnet. after that everything is mostly groovy but the individual components don't easily seem to be able to connect to a different vpn.
I remember having problems using tailscale vpn 24/7 and pihole on my home network with the phone pointed at the 192.168 address for DNS. Pages would take 5s to resolve and start loading.
Unfortunately, Pihole was less important than Tailscale and I have to put up with mobile ads.
I would love to know if there's a way to secure this though. I'm not prepared to have people constantly trying to login to my immich instance so it's only accessible via VPN
Bugs happen in all code. The difference is, anyone can fix stuff in open source. Closed source bugs are out of control and must be worked around. Usually by switching to OSS
The other day claude spun up 100 agents and took an hour to type 30k token document to tell me something was impossible to do. I googled it, found a pr on the 3rd link that showed it was possible. "You're absolutely right!!"
I wonder if this can explain something happening to me. If I select "random" at character select, I had a run of 30 or 40 where I never received the Silent. Defect seem to come up more often than it should, and Ironclad less often.
Claims of AGI imply that LLM's have intelligence. They don't, they are fancy probability machines. They don't THINK the way we do, they just do 200 matrix multiplications until their training data is massaged into what you need. They don't dream, they don't remember what you tell them. Even if you write one sentence, 'attention' means they will ignore half of what you say and key in on the wrong thing. This just happened to me today on a frontier model.
I'm not saying that AGI is impossible, but the focus on LLM's is probably not the right approach. I don't think we will ever make it until we understand the human mind better.
Current favorite language is Rust.