- nftables is default to `ip` family which only applies to IPv4. Setting it to `inet` will allow rules to apply to both IPv4 & 6; or `ip6` for IPv6 only. You can skip NAT rules, usually.
- dnsmasq: in addition to DNS and DHCP, turns on router advertisement with SLAAC. Some devices can get IPv6 address from stateful DHCPv6 server, others (e.g. Android) only work with SLAAC.
Random thoughts: one can get user's ssh public keys from GitHub on the fly (from `https://github.com/<username>.keys`), so that it requires a valid GitHub account to use this service, without (extra) auth process.
The time at which scalpers buy tickets becomes irrelevant. They can resell tickets at arbitrary-high-price + margin because buyer know they will receive refund for the exceed amount, thus pay only (final price + scalper margin) for the ticket eventually.
Bonus: buy ticket from scalpers after price settled, you will get a determined price, no more guessing and find inner peace.
The problem for scalpers is that if they buy too many tickets, the final price may become too high to be attractive for real buyers.
In the mug-scrubbing video, the person clearly pretends to wash the cup but does not seem to want to get their hands wet anyway. I'm curious as to when models can figure out that subtle thing.
Did you try scancode map registry[1]? I use it to swap Ctrl/CapsLock and it works well even with RDP softwares and run-as-admin apps which PowerToys couldn't handle.
Just watched a talk[0] about natural language understanding research in post-GPT-3 era. Old issues may has been solved, while new topics are coming to this area (quoted from the slides):
> then the amount of information we need to make billboards and signs make sense.
Subsequently, this applies to posters, letters, newspapers, and other types of text-heavy images, ultimately reducing the language modeling problem to an image generation problem.
OpenAI did exact the test for GPT-4. The raw, non-fine-tune GPT-4 is quite good at predicting confidence level ("highly calibrated" by their words). But the RLHF fine-tuning process seems ruin its calibration. Figure 8 on page 12 of GPT-4 Technical Report shows this dramatic changes before & after fine-tuning.