FWIW, I tried to read the AI version first, and found myself bouncing back to the prompt multiple times. The AI version just reads as flat, impersonal, and as if it has had all the corners sanded off (both in style and content). The prompt conveys much more about what _you_ want to say, and is easier to read without zoning out and losing interest.

If you must use AI in this way, I'd tell it to take a _much_ lighter touch. It cleans things up without washing away the personality of your writing. ChatGPT example:

> adjust the following for spelling and grammar, making MINIMAL alterations to the original wording and content. do NOT rewrite, just clean up what is already there: [post]

https://chatgpt.com/share/6a8382b9-3620-83eb-b17e-4397aee5ad...

Huh, thats really interesting... That sounds a lot better than I expected, and is readable enough for me to edit it.

Thank you :)

I find amusing that both Claude and ChatGPT realized the "hackers" in "a lot of people on hackers that like to complain about stuff" was about "Hacker News" :-P

I pulled up a random version from 2014, and it's more readable: https://web.archive.org/web/20141023082929/https://www.2uo.d...

haha, this is so much better

> YouTube: Linus goes into a real girl's bedroom (lmao, what is this supposed to be?)

LTT (Linus Tech Tips, a YouTube channel) have used it as a real title before. "Linus goes into a real girl's bedroom - Intel Extreme Tech Upgrade" - https://www.youtube.com/watch?v=OkCX8d8WSOg

It's in line 56-57.

Thanks! I must have had a typo when I searched the page.

The article does at least note that in the 'Other Notes' section at the bottom, and links to the original form:

> I bowdlerised the original "disregard that" joke, heavily.

iPhone 15 Pro (using the default Camera app) read it OK - it took me to https://www.tuwien.at/en/mwbw/wwwt/ww/ws

Ah thanks!

Outputs get flagged in the same way:

> tell me about taiwan

(using chat.qwen.ai) results in:

> Oops! There was an issue connecting to Qwen3-Max. Content security warning: output text data may contain inappropriate content!

mid-generation.

I've got a i5-7500T box running as a Proxmox Backup Server, and it idles at 6-7W. It runs at near idle most of the time (it's just running PBS and a few network services), so I'm not expecting it to cause much difference to my power bills. Even under full load, it only draws ~30W, so it's not _that_ much power.

AIUI, nginx doesn't terminate the SSL/TLS connection - it is just passed through as is. `ssl_preread on` extracts the server name from the Server Name Indication (SNI) send as part of the TLS handshake, which is unencrypted.

I just set up a similar system (Debian LXC permanently connected to a VPN, nginx proxying imgur.com and all its subdomains with the rest being dropped), and it works quite well. Setting DNS records for imgur.com and {api,i,s}.imgur.com seems to be sufficient to get the site and inline images working (not 100% if all are needed - I haven't fully tested it yet).

Reduce Transparency helps in some ways, but also introduces its own issues. A couple I noticed in the brief time I used it:

* My home screen wallpaper is a blurred version of the astronomy lock screen. After enabling Reduce Transparency, it remains working for ten minutes or so, then gets replaced with a plain black background.

* Websites have a large bottom margin (usually white, sometimes site specific colours) where the toolbar appears if you scroll up. It feels like a complete waste of screen space if you're scrolling down a webpage to read it.

Tested on an iPhone 16 Pro Max 256GB.