Interestingly Zelensky might be easier to deepfake than other world leaders because there's more footage and a wider variety of facial expressions to train on due to his acting career
Might not be so obvious to a soldier who has slept one hour a night for the past two weeks, eaten only canned corn and tuna, hasn't gotten ammo replenishment in days, and is watching this video on a tiny cracked screen with a dying battery with the volume very low as the internet connection fades in and out.
That's true for now, but eventually, as the technology progresses, it will take very little training data to create deepfakes of anyone unfortunately. Probably just a minute or two of video.
Is this true? I don't see much evidence that the technology is progressing in a direction where models will require less training data, if anything the trend seems to be towards models with higher and higher parameter counts.
Better models are coming out which are already pretrained on a significant amount of data, so the model already learned a lot about what is common to all example of video generation (keeping the edges aligned coherently at every frame, keeping texture and lighting coherent etc.) and will not need to re-learn that for every target.
Since initially, deepfake models were trained from scratch for every single target, you had to provide a lot of data from the person you want to target so that the model can learn what is common as well as what is specific.
Now you can get descent performance with much less data, since you only need to learn the specifities.
However, this only helps if you need a limited deepfake: The model cannot infer the exact facial expression of the target when they are, for example, laughing unless you provided an example of that in the training data (assuming there is no way to infer the laughing expression from someone by looking at other provided expressions). It will instead generate a generic laugh. All missing informations are substituted by what was seen, on average, in the pre-training phase.
That wouldn't work for a long complex deepfake meant to be sent to someone reasonably close with the target.
But for the types of deepfake where it's targeting a personality that we all know, but not very well at all, much less data is neeeded than before for a similar result.
At least in my experience - audio is much harder to convincingly fake than video. If you have heard the real person speaking, they have very specific and distinguishable patterns of speech.
You can fake it reasonably, but you need to have a very large collection of audio clips to do so, and if you do a bad job it literally jumps out at the viewer.
Video might be off, but it requires close attention and large screens to notice - much easier to miss if you're viewing on a phone.
This is called few-shot learning (or in the limit case, with a single example, one-shot learning). One way this may be achieved by first training a very general model (maybe with huge data sets), and then fine-tuning it into a specific example (this is called transfer learning). [0]
One reason researchers suspected this must be possible is that human beings, as well as other animals, can learn stuff by watching it for just a few seconds. But we have some prior baggage, because we spent our whole lives learning.. other, vaguely related stuff, and it turns out that knowledge is often transferable.
[0] This isn't the only way, there's also meta learning
I think the idea is that data needed to imitate a _specific person_ would require less data. Overall a model like that might have orders of magnitude more data in general and maybe a smaller amount required to imitate the features of a particular one.
I imagine it like - A master cabinet maker can make new variations of cabinets super easily once they've made hundreds of similar cabinets?
The real threat I think will not be in trying to trick people with fake videos, but instead deep flags: create fake videos showing awful things about the politician one wants to win, and then attribute the fakes to supporters of the guy you want to lose and make them look like animals.
Especially in coordination with a media campaign, it would be an effective way of undermining the opponent by making them look simultaneously grossly unethical and desperate. Even better if you can get the FBI involved to investigate whether the video was created directly by the opponent or "only" by his supporters. And anybody describing what actually happened could easily be straw manned into sounding like a nutter.
And the quality is already perfectly fine for this.
Not to mention, while it could be hard for random online trolls, certain state actors in your country of residence likely already have a ton of your face stored on video, celebrity or not.
That might be the best way to make him do something. When he realizes that people think the fake video makes sense, he'll claim the video was real out of spite and start defending a reasonable position.
He used to be a Democrat - one of the excuses people had for voting for him was that he would really govern from the left. There are (probably authentic) photos of him palling around with the Clintons. He praised Hillary Clinton's tenure as Secretary of State.
I don't think you can deepfake Donald Trump. The man is a recursive deepfake of himself.
Trump never came around to accepting Covid had to be dealt with (doing so would have easily given him a second term), and is just barely starting to condemn Putler. This idea that Trump masterfully chooses his positions based on popular effectiveness is highly overrated.
With him, everything was a calculation. Admitting COVID was a serious issue was thought to bring an end to the stock market boom, which was the big economic win for his re-election campaign. The market crashed anyway, before recovering ahead of the election.
In any case, 2020 might be the first time enough voters realized that a bull run on Wall St doesn't move the needle much for their own economic well-being.
I don't see how this narrative makes sense, and seems to be yet another example of "4d chess". The market promptly dropped ~30% in March 2020 in response to Covid, but was then inflated with trillions of dollars of newly printed stimulus money. Which, as an aside, has finally filtered into the consumer economy causing much of the price inflation we've been suffering.
Realizing Covid wasn't going to go away and coming around to addressing Covid in June 2020 or so wouldn't have affected the market, which had already priced in Covid. But it would have made Trump an actual crisis-time leader with the corresponding popularity boost just from leading people through a difficult time - publicly validating the hardships everyone is feeling goes a long way. But instead, he dug his heels in opposition to how things were developing.
If anything, I think saying that Trump is a basic contrarian is a better predictor of his positions. The problem is that while our society has major flaws that make contrarian viewpoints compelling, contrarianism on its own does not produce useful reforms.
> But instead, he dug his heels in opposition to how things were developing
Mostly in blue states! back then, red states where doing fine, but Seattle & New York were the first major epicenters. It was a deliberate political calculus to cast Democratic-led regions as inept. Florida "banned" New Yorkers for a bit. Minimizong COVID also made businesses happy, so it appeared to be a 3-fer for Republicans, until it boomeranged.
Still, acknowledging Covid as a real problem would have let him drive the density difference even more. It's harder to blame democratic leaders as inept if you are downplaying the seriousness of the underlying problem. Also by June 2020 it was quite apparent that the problem was not going to be isolated to nursing homes etc, even if the absolute numbers weren't bad everywhere.
> In any case, 2020 might be the first time enough voters realized that a bull run on Wall St doesn't move the needle much for their own economic well-being.
Wall Street was on a bull run leading into 2016 too, and that realization was one of Trump's arguments at the time. Once he won, he dropped that line, but if 2020 was such a time, I'd count 2016 too
I just want to point out that this is part of the premise of the original article:
> Harvard’s admissions policies make me especially queasy, because the burden rests so heavily on Asian Americans. Still, it advances the important goals of exposing students to a somewhat broader range of colleagues and helping to lift more underrepresented minorities into positions in the power elite, which remains heavily white.
Generally, going to prestigious schools benefits under represented minorities more than it benefits Asian Americans to an extent that affirmative action can be effective at addressing systemic inequalities, but it's still discriminatory against Asian americans to do so. It's a matter of opinion whether it's a good tradeoff or not
Is there anything that the admissions committee can do to make sure that admission is totally discrimination free? Even evaluating by pure test results would mean that there is systematic discrimination towards the unprivileged. Asians are extremely well represented at top schools, i'm not sure what they are fighting for?
The thing is that there's two levels of possible discrimination; systemic discrimination and admissions discrimination. Pure test results are discrimination free at the admissions level, but they don't address systemic inequities. To accommodate for systemic inequities, we use admissions discrimination. Again, that's not necessarily a bad thing; the author thinks it's a worthy tradeoff.
> Asians are extremely well represented at top schools, i'm not sure what they are fighting for?
The truth is that Asian Americans have to get higher test scores / GPA / extracurriculars to get into prestigious schools. It's true that this might have a positive affect on reducing systemic discrimination in the long run, but it's hard for Asian American applicants to accept that the burden should be placed on them.
Thanks! And that is so cool! It looks like the intended use case of this is to generate documents or short snippets of LaTeX, but not have to hard code the answers?
I wish I had something like this back when I actually wrote reports and stuff in LaTeX. Now I barely have any use case for my own calculator, except occasional unit conversions.
The use case that I had in mind for mine was electrical engineering students, where I had to often use complex numbers in polar form, in degrees, for circuit questions. My calculator at the time had no obvious way of entering polar complex numbers in degrees, I ended up typing things like `e^(i*angle/1r)`, where the r was "radian", and when the calculator was in degrees mode, "1 r" would evaluate to "180/pi".
It was also cumbersome to write long expressions for. It showed something like 24 characters per line, enough that I'd often get multi line expressions. These expressions would be full of brackets, and sometimes I'd hit enter and not be sure if I missed a bracket. (Hence my goal of having a LaTeX display to show you what you're entering).
Yeah, I made it to do my chemistry homework where a lot of problems were duplicated with different numbers. The truth is that it's not well tested outside of that though so it's hard to trust for complicated stuff. I might end up using yours :)
An interesting example is hollow knight. It’s one of the most successful indie games of all time and it’s almost entirely coded visually: https://unity.com/madewith/hollow-knight
Anecdotally it makes my fans spin really fast for a 2D game but that might be because of the fancy shaders
at first I was jealous of the GUI navigation, but then I realized I don't use any GUIs other than firefox, for which there are plenty of plugins.
The Vim emulation is pretty cool, the closest alternative I can think of is https://github.com/glacambre/firenvim, but I disabled it a while ago for being kind of clunky
You sound like somebody who's sampled a grand total of 1 tiling wm and now thinks they know the field. You do not.
By your reasoning, i3 is a copycat of ION or wmii and shouldn't have even bothered.
There is plenty of room in the market for more tiling wm's. Yes, you can technically run i3 in conjunction with Gnome but then you are essentially losing all the QoL and floating features of Mutter including the expose view which I use heavily.
I think many people haven't realized that different paradigms fit different people's brains in different ways; there are better paradigms for a specific person and task but it's impossible to generalize.
