Tech News Round-Up

If you care about security in the slightest, turning off Secure Boot—on any computer, PC or Mac—is a very bad idea. The real solution is to make (encrypted) backups and keep them up to date.

This is all pretty fresh because I recently bought a new PC and took a deep dive into Secure Boot, UEFI, GPT, and Linux support for Secure Boot via shim.

With T2 disabled, anyone technically inclined can fuck with your Mac, and anyone who gets your password (say, after fucking with your Mac) can decrypt your disk (you’ve enabled disk encryption, right?). Likewise, Secure Boot prevents anyone without physical access from trying to install os-level malware on your Linux/Microsoft box. Microsoft has required Secure Boot support from OEMs since Windows 8.

Unless you know exactly what you’re doing, keep T2/Secure Boot on.

More info on the SRM situation:

TL;DR: SMR drives definitely significantly underperform compared to non-SMR. However, they are still adequately speedy for many tasks. The logic in the firmware is pretty impressive at how much it can overcome the performance issues inherent in SMR. The problem is that WD is using SMR to cut corners on cheap/small disks and tried to hide it. If they used the technology instead to make very large disks, and let the customer know about it, it might be praised as a great product for people who need a lot of storage and are ok with slower speeds.

Data backups seem like a good usage for SMR drives, but I would like to see TRIM support on Seagate ones (I’m not sure what WD supports). I have several Seagate Expansion SMR drives, and after you fill them up once, you see drastically reduced speeds from then on ever after deleting files. I’ve seen speeds lower than 10 MB/s after you fill a drive and try to write to it in bulk.

How are you writing to a full drive? It’s full!

I meant after you fill a drive with data and delete things. Since there is no TRIM support the drive still has to shuffle around unallocated chunks of deleted files, slowing down writes.

Ah, I see. The thing is, no matter what kind of drives you have, filling a drive is always going to cause problems. I always make sure to have enough free space breathing room. If it starts to get tight, I expand the storage. So I guess even if I used those drives, I would never encounter that problem.


1 Like

I’m surprised it only took 186 reels of microfilm to store 21TB of code.

If i remember right photo reductions made great strides. I doubt they are conventional micro form especially if the images in the article are accurate. Could be thousands of meters worth of film in one reel as well…

How about that meow situation?

Having an unsecured database connected to the Internet in 2020 is like leaving your wallet in an unlocked car with the windows down parked in a busy downtown.

1 Like

I don’t even know where to put this one.

1 Like

Guess leftists really love their trains.



Come on antitrust regulators where you at?

So Fawkes just came out with version 0.3. The tl;dr on it, is you take a photo and run it through it and it inserts noise imperceptible to the human eye but devastating to machine learning algorithms trying to build a dataset of pictures of you.

Pretty cool and an example of that concept I’ve been seeing a lot where we’ve spent a lot of time getting ML to work with perfect datasets. It’s gonna be a long time coming before we are able to handle adversarial datasets like this.

That said I downloaded it and am running it on an image of me I took in college. It took a while, like a few minutes. That said side by side they look the same.

So I guess the idea is run all photos that you put online through this and new photos of you won’t match as you if you do this long enough.

Kinda cool.

Edit: In the interest of keeping detail, while the thing runs it happily maxes out your cpu. So if you were to actually try and like do this to all your photographs, I’d maybe set up a pipeline that does it on another machine. And it’d be rude of me to not give examples so I did it again on a photo of an actor I had lying around:
This is the original

This is the one manipulated to not be recognizable to facial recognition software:


Basically YouTube is getting rid of the ability for people to suit subtitles to videos, which is part of a bigger problem.

Video owners can still create and apply subtitles. They only got rid of open community submissions.

The Linux kernel patch submission method, then.