I’m not even convinced the audiod thing is Regin; whatever is going on is way less sophisticated even based on what the OP posted from volatility. I don’t think the hash they gave vx-underground is even from the sample from the original screenshots.
I think this person is just karma/clout farming badly and the screenshots are of some even more basic RAT.
He hasn't actually confirmed that the image he's processing is recent or if it was a test image and by "I found", he means he was able to find the thing that was known to be there. The Twitter thread has some people asking for clarification and none have been received yet.
Hard to say because I'm still stumbling trying to remember a chord most of the time. I'd say 20 wpm when the stumbling is only moderate. I need to get that muscle memory trained!
Did you follow some convention for your chording or make something up yourself?
I kind of wonder if some layout that mimics wasd but uses the thumb buttons to indicate which “row” you are in could be intuitive to people who learned to type conventionally. (The intuition here being that most of us aren’t going to become keyer experts).
No, I didn't experiment with modes almost at all. I had one mode where I mapped the arrows to individual keys but in the end dint't use it - it's faster to enter a chord - especially chords for Ctrl+Arrows are nice.
For a time I made the mappings a little more memorable by forcing two related keys (like a and ą or o and ó) to have their chords differ in just one finger position - and that did work but it lowered the "efficiency estamates" of the generated layouts. In the end I reserved one thumb position for my custom shortcuts and allowed the optimizer to go crazy with all the remaining chords. After playing with both styles I prefer the latter. Entering text feels more a little fast-paced maze solving game where you have to figure out which fingers to move to transition between chords.
Interesting. As a vim fan I think I would be very unhappy with any layout that didn’t have hjkl as my home position. But, of course, the ability to experiment is a huge strength of open source projects.
What a cool project. I grew up playing with modeling clay, but never did anything with those skills. It is fascinating to see them used in something useful like this.
Maybe a scanner of some sort is needed, to share 3D printable versions of clay objects, haha.
In my setup I use Colemak DH mod which loses the Vim arrows but I added a modifier where the 'a' key (left pinky on home row) when held down switches the right home row to arrow keys. Hasn't been an issue.
Per Discord's press release, it appears only a small subset of photo IDs were leaked:
>The unauthorized party also gained access to a small number of government ID images (e.g., driver’s license, passport) from users who had appealed an age determination.
You seem to be reading the press release language exactly as they'd like you to read it.
Users only upload their government ID to Discord when the "Face Scan" [0] incorrectly estimates their age as being less than 18. Discord could reasonably classify this as a "small number" of users who need to upload their government ID image. That wouldn't preclude it from also being every user who needs to upload their government ID image — unless there is some other system that also requires them to upload it?
With that in mind, here's a rephrasing of the same statement:
> The unauthorized party also gained access to all uploaded government ID images.
Their press release does NOT say it's a small subset of photo IDs. It says a "small number" of government ID images — nothing about percentages. This would be consistent with the "small number" of users who need/choose to appeal an incorrect age estimation from Face Scan.
This comment is a fantastic study on how to adversarially read press releases like this. I suspect it's exactly correct: likely all photo IDs were leaked, but they decided to cast it as a small number by implicitly comparing it to the number of all Discord users. I guess we'll have to wait and see if that's actually correct. We may never find out.
There are two options for verifying your age on Discord - face scan OR uploading government ID. So some people may have uploaded their ID instead of doing the face scanning, for whatever reason.
For example if the face verification failed and you need to file an appeal which requires uploading government ID. That's likely sizeable number of users, especially since the breach happened shortly after the requirement was implemented and many existing users had to do it.
I'm not sure why this is being downvoted. Commenter is entirely correct. If someone has an answer to their question that would add credibility to Discord's phrasing and GP's interpretation, I'm all ears, but otherwise it does seem like this is the case, and every ID they've collected has been leaked, not a subset.
(To say nothing of... does it matter the amount of IDs leaked?)
If a message like "I'm 12", regardless of context is reported, Discord will ban the account & hold it hostage until user sends selfie + ID to them via support. (the compromised portal, not a third party app dedicated to this)
They intentionally chose NOT to disclose a date range or even how many ID tickets compared to standard tickets were leaked.
Oppo uses these in some of their phones. They gave a factory tour to the "Know Art" Youtube channel, which made a good video on it: https://youtu.be/qAZ-q3KmDHM
Right, but they didn't use rented GPUs, so it's a purely notional figure. It's an appropriate value for comparison to other single training runs (e.g. it tells you that turning DeepSeek-V3 into DeepSeek-R1 cost much less than training DeepSeek-V3 from scratch) but not for the entire budget of a company training LLMs.
DeepSeek spent a large amount upfront to build a cluster that they can run lots of small experiments on over the course of several years. If you only focus on the successful ones, it looks like their costs are much lower than they were end-to-end.
If by "they" you mean DeepSeek, they're not saying this, since you might not actually be able to rent a cluster of 512 H800s wired together with high-bandwidth interconnects at that GPU-hour price point. If you rent smaller groups of GPUs piecemeal in different locations and try to transfer weight updates between them over the internet, it'll kill your throughput.
For others who run into the same problem, the file can be accessed via https://fabiensanglard.net/gebbdoom/index.html#:~:text=High%... . (I've highlighted the link to click.)
reply