I’ve also become more and more convinced that these colossal huge general models are not necessary for many useful tasks and smaller more focused models will do just fine for a lot of the workloads people will actually be willing to pay for. Particularly once these companies have to stop selling at a loss.

I like and use Cloudflare but it’s worth mentioning you must use their dns for domains you purchase from them.

Generally when accuracy matters I ask for a direct link with proof and for the specific part of the doc that is related. I either get exactly that or it’ll realize it’s hallucinating and can’t prove anything.

Hugely wasteful but unfortunately still faster than how unusable search has become.

Changing the password on my password manager is so stressful.

Legit have to plan a good time of year to upend my life like that.

Had to Google it but it looks pretty nice. I thought you meant that you were injecting giant ads into Android TV.

I can’t imagine what nightmarish vulnerabilities Microsoft knows about and is hiding because they would require too much effort to patch. I bet there are some really crazy things that have probably been wide open for decades if you only knew where to look.

Generally people should be more worried about the Ring camera across the street than someone using AI to analyze signal shaping data from the Wi-Fi to figure out what room you are in.

I’m interested to see what kind of hardware will produce “good enough” AI capabilities in a couple years as things are refined and tuned further. The gap between the absolutely massive commercial models and open source models keeps shifting but I don’t have the same fear that I had a few years ago that it might not be possible to get good results from anything less than millions of dollars worth of hardware.

Buy the game after a few years of bug fixes on sale with all the DLC for basically the same price as a sandwich… or pay $80 for a buggy broken incomplete experience with no real guarantee any promised content will ever materialize.

Although I guess I need SOME people to keep buying them at launch to subsidize my frugality.

I deployed Technitium with docker, but generally this got me heading in the right direction with the initial setup. It’s more of an overview and quickstart than an in depth guide though.

I was doing basically this with a different sync tool, but I had a couple issues with it:

• If I needed to toggle ad blocking I had to go toggle it on both pihole instances
• When troubleshooting an issue I had to check two log sources, or disable one and then reproduce the issue again
• Sync was one directional so I could only make configuration changes from the primary instance

I’ve been a very happy Pihole user for years and years and Pihole 6 is the best yet, but once you’re dealing with multiple pihole instances, Nebula Sync and Unbound, then Technitium is actually simpler to manage since it does all that natively.

DNSWeaver has support for caddy labels too! Specifically for use with caddy-docker-proxy. So yeah, really good fit for your architecture.

Oh do I have a treat for you, check out DNSWeaver.

It’s designed to do exactly that, to automate creation of DNS records for container services. I use it with Traefik. It reads from the same labels that Traefik already uses to proxy services but if you already use another reverse proxy and don’t want to switch it supports dnsweaver-specific labels as well which are easy to add to your current deploys.

I used it both with pihole and technitium and actually used it to make the migration easier. Great tool.

People forget that social media is escapable and you can just leave.

Switching SSH to a non-standard port can cut down on log noise but it doesn’t really help with security. It’s trivial to identify ssh running on any port and attackers typically do full port scans anyway.

I’d put that effort towards allowlisting only trusted public ips or setting up wireguard/tailscale for ssh access instead.

Crimes only count for poors.

I migrated from pihole to technitium a few weeks ago and it was so smooth.

Native support for clustering is huge. I didn’t even realize how complex managing the pihole had gotten trying to get it to sync to multiple instances.

I back up to local storage and then replicate offsite to S3 nightly.

On-prem backups are great and cheap and fast and definitely plan A but a robust backup solution is going to require offsite storage of some sort. Object storage is one of the cheapest ways to do that for most situations, particularly for things that can’t be replaced like photos.

Backblaze B2 is about $7 a month per TB.

Almost every major backup solution natively supports S3 compatible storage.

This looks like someone was afraid they might accidentally cook something.