This is basically Claude Code with less extra steps

It does not follow that merely looking at or altering the generated code makes it acceptable. By the rules that have been adopted, a contributor cannot take substantial implementation produced by an LLM, clean it up manually, and then treat the resulting patch as if it had been originally written by them.

I wonder how they intend to enforce this one. Raw LLM output will trigger some kind of “Spidey sense” for anyone woo has seen enough of it. But I don’t know how I would tell manually cleaned up LLM output apart from manually written original code

I think your missing the point about the whole “linux desktop customization” hobby. The fun is in building the thing yourself by assembling software that “does one thing and does it well”. Getting to use after it’s built is just added bonus :)

Yeah, I could see a credential stealer being smuggled into an alternative frontend…

It also made my browser freeze for a moment when I first found and loaded his page before I blocked all JS from his domain. Not stealthy at all… Many nerd (like most of us lemmy users are) will instantly open devtools when a random page that’s supposed to be just text freezes the browser for a second.

Like, c’mon man, setInterval with even 10 ms of delay would have made the attack both stealthier by not insta-freezing the brower on load, and more efficient by getting way more than 2000 request out of each reader…

Also, their website seems to be down now :)

I did not follow this drama, and I don’t know about this list everyone is mentioning. Is this something that hides specific users posts and comments from the Tesseract UI ?

“Apparently only three people who aren’t me actually want a place that’s not just a circle-jerk of rage and hate.”

I think this is the best part of their message

I never knew this fun fact ! I’ll try to use it with my students, and see if I this gets me more attention when I give them the trick to compute powers of 2 in your head (2¹⁰ =~ 10³, which leads to ln(10³) =~ 10*ln(2) ).

For anyone wondering, the trick is : Powers of 2 below 10 are easy enough to remember or just compute in your head. So if you want to approximate 2^x, write it as (2^10*a)*(2^b), with a=x//10 (integer division) and b=x%10 (modulo). b will be 9 or less, so 2^b is easy to compute, and 2^10*a = (2¹⁰)^a =~ 1000^a is also easy to compute.

For instance, 2³² = (2³⁰)(2²). Since 2³⁰=~1000³ ; and 2²=4, then 2³² =~ 4(1000)^3 = 4 billion. Once you get used to it, you’ll be able to say that 2⁶⁴ is approx 16 billion billions in no time ! :)

Me, after I did sudo rm -rf /mnt/chroot when my whole home directory was still bind mounted inside the chroot…

This was before ChatGPT, guaranteed 100% natural stupidity

I’m in France timezone so hopefully we can find a time of the day that works for both of us, but feel free to DM me here or on Matrix (@pcouy:matrix.pierre-couy.fr)

I was about to type pretty much the exact same thing, but I figured it would already be here for me to upvote :)

The ansible role I made does not cover this. If you make a server an exit node for a network, the server gets configured to forward everything that comes from the wireguard network. Since my use case does not need this, I don’t think I’ll add it, but the role could be expanded to support ACLs by taking inspiration from how it deals with port forwarding.

Do you have specific examples of the “different firewall requirements” ?

I have a custom made ansible role that does what you are asking for. I’ve wanted to clean it up a bit before putting it up on Github, and your post just gave me a motivation to stop procrastinating it :) This was initially made for my own use, and I’m not an Ansible expert so it can have some rough edges, but I tried my best to make it adhere to the Ansible way of doing things (applying the same role to a host twice should be idempotent).

Using this role, I’m able to quickly define wireguard networks, define which peer participates in which network, optionally configure a peer as a router (with NAT and port forwarding) for a given network. For Ansible-managed hosts, it installs wireguard, generates and deploys wireguard configurations, and enables IP forwarding on exit nodes. It can also generate ready-to-deploy wireguard configurations for externally managed hosts.

The repo is at https://github.com/pcouy/ansible-wireguard . I tried giving some instructions in the readme but they may not be detailed enough. If you try it, please open an issue for anything that’s not clear enough from the readme so I can improve it. (I tried making a “Quick start” section for people who never used ansible before)

As someone who teaches CS and grades assignments, the last few years have been really rough. Academic dishonesty has skyrocketed with models becoming smarter and students becoming more dependent on them. Any assignment that’s above average will make me suspicious, and when it appears to be 100% AI generated, the feeling that I spend more time grading than the student spent working on it is awful. Even when I’m almost sure a work is AI generated, unless there are some dumb leftovers such as “As an AI assistant […]”, I can never be 100% sure. This causes me a lot of headaches because the only thing worse than rewarding dishonesty would be not appropriately rewarding an outstanding assignment.

As much as I’d love to have à software tell me with 100% certainty which parts (if any) of an assignment are AI written, AI detectors are all snake oil, no exception. They exploit teachers’ helplessness to make false promises that we really want to believe in.

Moreover, I don’t think fully banning AI use is a sensible thing to do. LLMs are a thing, whether we like them or not, and using them in a sensible way is a useful skill to learn. There’s one big issue though : on one hand, assignments are made so that the problems students have to solve all have well-known solutions. This is required to make sure the assignment is doable in the first place, and that teachers will be able to help. On the other hand, LLMs are disproportionately good at classic assignment problems since there are so many published solutions online (which then end up in training datasets). Moreover, assignements are usually made to guide students through a larger problem by breaking it down into smaller problems, which is basically the perfect prompt for a LLM. This means students can get away with the laziest uses of LLMs (which usually won’t work with real world problems). In the worst cases, the only “skill” some students learn is to throw a PDF at whatever AI they paid for, ask it to solve the assignment, and copy paste the output without thoroughly reading it first.

Teachers clearly need to adapt. There will always be a few students who fail to learn in every class, but when so many students don’t learn, it’s the teacher who is failing to teach.

I did something similar for digitizing old camcorder family videos. I hooked up a VCR through a cheap elgato RCA -> USB capture card to my home server, and did something similar to you with a preview feed available on my LAN.

I even built a (really ugly) web UI which lets me pick an audio and video source, start the preview stream and display it in the browser without actually starting the capture, then start/stop the capture, edit the ffmpeg flags, and set the output file name. I can share it with you if you are interested

Are you the author ? Anyway, thanks for sharing, that was a very interesting read

I don’t get why anyone would downvote you. This is a very legitimate question.

As an AI that is not an LLM, it cannot write or talk

Thanks for the links. I found out about suckerpinch’s video a few days ago. Really interesting how he found an heuristic to auto-make reward functions. This is almost orthogonal to what I’m doing, and could be combined with most RL algorithms.

I’ll make sure to check out the paper as well

(Bonus question : do you have any idea why I got down voted so hard ? I thought this is perfectly on-topic for the community)