HN user
More commonly, voltage glitching does rely on removing capacitors etc., and extremely precise timing (nanosecond resolution, or below).
However, the fun part about this particular attack is that it does not require quite the same level of precision. The voltage is dropped incrementally until the CPU starts to make faulty calculations. It just so happens that AES is one of the most complex operations, and thus, the first to start faulting.
Gaming performance usually comes down to graphics drivers.
I'd much prefer a slightly dry but genuine personality, than a hyperactive and fake "youtuber personality". The content speaks for itself. He generally takes safety pretty seriously, too.
This requires the assumption that storage will continue to exponentially (or at the very least, linearly) decrease in cost. It also requires a certain amount of good luck. I would hope the IA has a reasonable amount of data resilience, but you never know.
A list of similar sites:
I used to have access to papers via my university, but I never used that access. The UX of scihub is infinitely better. Anything else just added unnecessary friction to my research process.
This post is still getting linked from other places, so I think it's helpful to point out that it's almost certainly fake. (Hence its [flagged] status)
The images shown do appear to be adversarially generated inputs against some NN-based image hash or classifier, but there is no evidence to suggest that this is at all related to Apple's NeuralHash, or that the colliding hashes are from a real CSAM database (the target hashes are not public).
OP claimed they would "release 5 pieces of proof in the next 5 days" [1], and guess what, 11 days later they still haven't.
Look at OP's post and comment history, it's quite clear that they are a troll.
In the mean time, it has been actually proven that hash collisions against NeuralHash are trivially possible, see [2]
This is a lame attempt at trolling.
Going horribly wrong is somewhat typical for large software projects.
Codebases with high levels of abstraction and metaprogramming are often easier to understand in IDA/Ghidra, if your goal is to understand how a specific part of the program works.
"2 minutes setup" is mentioned in various places on the page, as well as on the chrome store page.
Why 2 minutes? I assume this is used as a stand-in for "very short time", because there's no way it actually takes two minutes to install unless your internet is particularly slow, and I don't imagine there's much to configure.
It seems a bit out of place, especially when the extension itself centers around the problem of 5 seconds feeling like an eternity.
Edit: I just installed it as a test, and it was ~instant.
I think that's a big improvement, thanks!
I tried one of the example queries, "UDP".
Out of the 45 search results, the most relevant one (the UDP RFC itself) was all the way at the bottom, listed as "RFC0768" and tagged as "legacy". (I'm not sure what legacy actually means here?)
Anyway, I clicked the link, which took me to a 404 error page - https://www.rfc-editor.org/rfc/rfc0768.html (The correct URL is apparently without the leading 0)
This site does seem genuinely useful, but I think the UX could really use some polishing here, especially since I literally followed through with one of the example queries in the middle of the homepage.
While it's certainly aesthetically pleasing, personally I find the link text too low-contrast. I think it should be slightly brighter.
I normally prefer and use dark-mode UIs, but this one is actually slightly harder for me to read.
Some people find them motivating, and a useful guideline for tracking their progress.
Personally I find them paralyzing - so I'll just progress at my own speed - nobody is forcing us to use the timings proposed in the repo.
we have the clout to send Medium a message.
I'm not sure we do. Medium is comparatively a much bigger and more mainstream platform than HN.
Just because it appeals to nature, doesn't make it good.
While I agree it was likely lighthearted, that does not exempt it from criticism.
I don't take issue with anything listed in your comment. The comment I was originally replying to described:
a) Bypassing their own code review processes.
b) Creating a contributors list, just so they can put Carmack at the top of it (what about all the other contributors?!)
IMHO this crosses a line. Not in a big way, but one worthy of comment.
So if people wish to be treated like normal human beings, they must hide behind pseudonyms?
That may be the current reality of the situation, but it doesn't mean we have to keep it that way.
I'm not saying he doesn't deserve his status, I'm just saying let's not get too fanatical about it.
John Carmack is awesome, but I find this level of deification (of any individual) kinda creepy - it just feels unhealthy. Would you like to be treated that way, in Carmack's position?
I realize that your comment was likely made in jest, but it still bugs me.
Who are you quoting?
At time of writing, pornhub.com has 69 referrals - I suppose someone was just having a laugh.
ARM has been around for decades now (and ARMv8 for one decade). Anyway, if nobody jumps to new architectures, how will anything get ported?
You've shortened 6,463,545 links so far. Assuming you're using a charset of [A-Za-z0-9], that gives you 14,776,336 total possible 4-character URLs. Almost half way there!
What's your plan once you reach the 4-character limit? Roll over to 5, or something fancier?
It's an impossible problem to solve completely. The most trusted of reviewers can still be bought out, for a price.
I think the best way to improve things is via increased transparency - If I can read a reviewer's past reviews, I can make a better-informed judgement of their reviewing integrity.
The problem is not Ads themselves, it's the fact that they inject perverse incentives into the entire tech ecosystem.
To maximize your advertising revenue, you need to track your users as effectively as possible. This:
a) Reduces user privacy. Even recent developments like FLoC, which appear to be pro-privacy on the surface, are really just yet another datapoint with which to violate the privacy of users.
b) Reduces performance. It's easy to blame trendy bloated tech stacks for the state of the web, but the reality is that a big chunk of the slowness comes from ad-related tracking and delivery technologies (install privacy extensions on a low-end system and see the difference!). This reduced performance disproportionately affects those with lower system and network resources, further reinforcing global inequality.
Although some will disagree, I do think that it's possible to advertise ethically, but no large corporation operating in a capitalist society is going to be doing that voluntarily - unless strict regulation comes down from above (Spoiler Alert: It won't).
All that said, I don't blame the author for working in adtech, but perhaps only due to my rather bleak perspective that we're all just cogs in the capitalist machine whether we like it or not.
Don't hate the player, hate the game.
P.S: Cynically, I think the author's announcement of their charitable donations counts against them. It makes it seem like he's trying to "offset" the harm caused by his work, even if he won't openly admit that such harm exists. Charitable donations may make you a net-virtuous individual, but they do nothing to address the harm of adtech itself - which is orthogonal to the author's original claim that they believe that advertising is a good thing.
So we train an ML model to convert images to text. How do we encode that text information?
you simply have a tool which extracts the text, transforms it however you please, and then puts it back as an image.
This is extremely non-simple. It is a strictly harder problem than just... encoding image+text in a sane container format like HTML or SVG.
Seems a bit redundant, if the page is being served over HTTPS, which ultimately relies on DNS records.
The advantage of PGP is that it can be verified entirely out-of-band, and distributed widely.
It's a public key, you can perform the "brute force" (factorisation) entirely offline, to derive the private key. Hypothetically. For now, RSA-1024 is too expensive to crack, for mere mortals.