There’s likely been trillions of posts on Facebook during that time frame. Is a sample size of 8000 really sufficient for a corpus that large?

I assume you mean the date and time settings page. This is KDE’s one:

I have to imagine it took a lot of work to verify conclusively whether something was or was not generated

The study is by a company that creates software to detect AI content, so it’s literally their whole job

(it also means there’s a conflict of interest, since they want to show how much content their detector can detect)

It’s a much larger sample size than a lot of studies.

It’s an extremely small proportion of the total number of Facebook posts though. Nowhere near enough for statistical significance.

This doesn’t have anything to do with encryption. They had a public database (anyone on the internet could query it) and forgot to put a password on it. It really shouldn’t even be public.

In that case, how/why did they only choose 8000 posts over 6 years? Facebook probably gets more than 8000 new posts per second.

You don’t need Docker. The agent is a single executable file so you could just manually copy it to /usr/local/bin if you wanted to.

I’m working on making it easier to install on Debian systems by creating a Debian package (and eventually a repo): https://github.com/henrygd/beszel/pull/497

How would they even attempt to implement this? Will the US end up with a Great Firewall like China has? Even if Chinese models are delisted from Hugging Face (since Hugging Face is a US company and has to follow US law), they could just be hosted elsewhere.

They’re different things. The comment you’re replying to is talking about different search engines. SearxNG is a metasearch engine - it combines results from multiple different search engines together.

Both are important. If you use SearXNG but only enable Google, it’s essentially just a proxy for Google and you miss out on most of the value of it. Some of the other search engines may have better results for some searches than Google does.

Most of the systems where this worked did it for any unrecognized words.

I suspect that most usage of the model is going to be companies and individuals running their own instance of it. They have some smaller distilled models based on Llama and Qwen that can run on consumer-grade hardware.

I’m considering placing a bulk order for these stickers: https://cnliberalism.org/store/p/my-tariffs-did-that-sticker

One of the main features of Deepseek is that you can run it yourself. It doesn’t matter if Deepseek are based in China if you run the model on your own servers and thus guarantee that your data doesn’t leave your own data center.

No I’m trying to use it with my work phone. I can’t connect my personal phone to the work wifi since it needs a certificate (802.1x) but there’s a separate guest network I can use if needed. Guest network is entirely isolated - different hardware, different backhaul, different IP range.

I love KDE Connect but I can’t figure out how to get it to work at work. Probably some firewall thing. It works fine at home, but can’t find my phone at work.

Why not Matrix via Conduit?

Not sure where you got that number from, but it’s way off. Average revenue per person is $12.29 per quarter (https://s21.q4cdn.com/399680738/files/doc_financials/2024/q3/Earnings-Presentation-Q3-2024.pdf slide 11)

$18/month would be a good deal if they still had all the content they used to have, but they’ve been removing content in favour of their own shows for a long time now. It sucks. You used to be able to find practically anything on Netflix, and if they didn’t have it available for streaming, they’d lend you a DVD (included in the subscription price).

That and the $18/month tier only goes up to 1080p. In 2025, you still need to pay more for 4k content?? It used to be 720p though, so I guess it’s not all bad.

100% agree. Note that some cheap VPS providers are single-homed (only have one internet connection) with a budget provider like Cogent, but the good ones are usually multihomed.

Hetzner are great. One of the providers I use (HostHatch) is trying to have pricing similar to Hetzner, but in a larger number of locations. Their sale pricing during Black Friday is even better though.

Yeah it’s part of their overall strategy to be seen as a core part of the internet / the web. Same as Yahoo in the 90s and early 2000s.

The more people that use their free services, the more appealing they are to advertisers compared to competing ad platforms (broader reach), and the more paid subscribers they get.

Products like Visual Studio, some Jetbrains IDEs, VMware ESXi, and a lot of SaaS products, are (or used to be) free for individuals or for open source usage for a similar reason - people get familiar with them at home, and end up recommending them and buying them at work. A few individuals liking the product can result in large companies signing paid contracts for tens of thousands of users.