Jnr

All good points, but I am not sure who I trust less. Chinese companies or random people on Hugging Face. :) It is still a black box. And I have a feeling I could not abliterate those models on my own hardware.

I am not saying I would not use any of them at all, but I would avoid doing certain tasks with them. And with the testing I have done with abliterated models in the past, it is not 100% successful and it can still hit the same walls as original model.

No political censorhip? Try asking Chinese models about Taiwan.

Some censorhip and bias towards certain ideology is baked into those models.

While it is not a problem for programming tasks, I would never use it for certain other tasks.

I am even more surprised about the multiple down-votes for this comment just because I like some feature. Must be a bunch of grumpy old keyboard warriors on this thread. 0__o

Surprised to see so much hate towards passkeys. I absolutely love them when I can use them with Bitwarden and my own SSO. Sucks when I am limited to hardware tokens like yubikeys.

Not sure what you implemented.

They provide several IP blacklists. None of those seem to be false positives. You can also add custom 3rd party blocklists.

They also provide several different rulesets. It is up to you to choose which ones to use and fine tune. LLMs can be very helpful with that.

And there are 3rd party dashboards and tools that help you manage it more easily.

I use the free version as a simple WAF on multiple servers and it blocks a lot of bots. It did require some initial finetuning though.

Are there any better open source solutions?

I have been on the same Arch installation for 10+ years and the previous install on a different device was also ~8 years old. Could have just transferred files over and be done about it but I wanted to refresh my knowledge. :)

Kubernetes supports multiple pluggable container runtimes per node. Can be quite useful when requiring some virtualization or running certain docker-in-docker scenarios for CI/CD workers. Maybe it can be done with Docker as well these days, but with kubernetes it is out of the box support.

Perhaps you could manually set up some VLANs and set up Docker networking to use those, but that has to be custom setup, none of that is a standard feature of Docker.

So instead of having to reinvent the wheel for each of those problems, you can just take kubernetes instead. And while it is a lot to learn, LLMs can make it a lot easier to do. For years I was running docker-compose stacks for self hosted things, but I replaced that once LLM agents became a bit less stupid. :)

Having separate bare server for each service would become quite expensive.

Yes, kubernetes makes it much easier to configure proper firewall through network policies.

Docker has networks but there is no good options to limit it beyond those and it is extremely basic and basically leaves everything open between your proxy and app.

In the age of LLM agents, still going for Docker/Podmam on anything besides local development box seems like a weird choice.

Docker has poor tooling for network level security between the containers, has issues with different runtimes per container, etc.

It is just a bit primitive if you want to expose multiple services to the internet on the same server. One of those apps will get compromised and then all the others will follow.

If you want a decent self hosted server, ask your frontier LLM agent of choice to configure kubernetes (on something like k3s) with mandatory userns mapping so nothing runs as root on the host, default deny firewall so inter-container communication is as locked down as possible, and if your router supports, set up VLANs so none of the containers can access your other devices on the LAN. Use something like backrest to handle backups, alertmanager and Grafana for monitoring, Keel for auto updates. Also consider separating ingress for public and internal services and use Tailscale with split DNS to acces the internal entrypoint. Set up Crowdsec as WAF and subscribe to their free blocklists to filter out bots. Ask it to set this all up using Ansible, so it can be maintained.

While this would be extremely time consuming to set up and maintain by hand, an agent can do and test it in a few hours.

Because it still doesn't fully support all of the hardware on M1. I know it because I used Asahi while I had M1. Now I have M4 for work and it doesn't support Asahi at all.

I just use Linux PC as much as possible instead. And that 4$ M4 is just a fancy presentation computer I use when I travel.

Personally I would spend thousands on an actually great Linux laptop, but there are still none around.

I bought my first minisplit ~4 years ago since none of the contractors were available in the hot summer. I bought the tools as well (vacuum pump, manifold, pipe cutters, etc.).

It cost me ~650$ total.

The next time I bought a minisplit (3.5kW one), I already had the tools and it cost me only 450$. A proper one meant for use during harsh winter would probably costs a bit more though, at around 1000$.

So I have set up a few of them for my family and friends so far. The units come pre-charged with gas for ~5M of ducts, there is nothing hard to set them up.

I also set up ESP32 controllers so I could attach them to Home Assistant, bypassing any Chinese cloud apps, etc.

With how bad Opus 5 and how expensive and limited Fable is, I would not be surprised.

I did and then rolled back. Gitea is far ahead in CI/CD features, which is the main thing I want from my own git forge. And they also seem to keep adding more meaningful features, not just refactoring to keep some federation thing working.

I block known DoH servers on my lan and forward all dns requests to my dns server.