Yes people are working on it. The thing you’re missing is that many of the contexts where the money actually is being spent is not really relevant to the public discussion around AI generated content. It’s more about making sure that nobody gets a bank loan using an AI generated voice and face, or that people don’t get scammed by a deep fake of their relatives, or that your government office isn’t being slammed with subtle propaganda, for instance. The trick to your concern is to change your expectations of accuracy. Flagging something as fraudulent with an ML is not treated by these systems as if it’s actually being fraudulent.
Imagine instead that you’re sitting on a swivel chair and you fire a gun with your right hand. This is more akin to how ejection of materials work. When you put your hands out on the swivel chair, the hands are still attached and are forced into orbit around your torso, when you eject material this isn’t the case, and it leaves the system.
I’m not sure why the idea that the cars will talk to each other is so “moonshot”. We have much more impressive pieces of infrastructure in place already. What might be moonshot is my proposal that it should be run by the US Postal Service.
I think it's a moonshot because it expects that we will have shared, open standards and protocols for all of this, which will either need to come from industry, or from regulators. Neither feels very likely. But then again, it could happen :)
They can't even get a standard way to cast your phone onto the screen in the car. It's just a fucking touch screen, and we can't figure that out because Apple and Google just can't let some portion of the money go and make their user's lives better. They have to have all the money or none of the money, and if they can't get all of it, then fuck you. Car manufacturers also won't even put in any sort of standards just so that I can upgrade the terrible stereo that comes in even the nicest of cars. God forbid they put a 2 cent RCA jack on the back of the stereo, maybe give you a read-only connection to the CAN-bus to display some data, and a rectangular hole in the dash (where the ugly ass screen is going to go anyways).
Location: Philadelphia, PA
Remote: Yes
Willing to relocate: Yes but only to New York City
Technologies: Data Science
Resume/CV: https://coen.needell.org/files/cv.pdf
email: coeneedell@gmail.com
I'm currently a staff researcher at Penn. I was trying to get a PhD but failed to secure offers, so I'm moving to industry. Historically I've been the sole data scientist and technologist on all projects, and as such have had to learn and become proficient in every step of the data science and analysis pipeline. I'm quick to learn, and I love learning how to use new tools, understanding their philosophy and what they're good for (or not good for!) I generally follow the "Data Science at the Command Line" approach to my work, and I like to build web systems for sharing my work from the lower level (d3.js over full-fat systems like observable, tableau, etc).
This argument seems to have an unsavory consequence. There’s nothing wrong with excluding and persecuting minority groups because they’re not part of the nation. With a country like China, where the now-majority Han culture colonized the minority Wu, Canton, Hokkien, and various small western cultures in the last 100(ish, it was going on before the PRC as well) years, and a narrative of national unity and shared history was fabricated by the government, can we defend anti-immigration policies on the same grounds?
>This argument seems to have an unsavory consequence. There’s nothing wrong with excluding and persecuting minority groups because they’re not part of the nation.
Minority groups have been part of different nations for millenia. So saying "they're not part of the nation" is not a given. They're just not be part of the majority group of a nation. Some nations might not even have such a majority group, but be composed of two or more different groups of roughly equal size. Nations aren't monolithic things. But that doesn't meant they're "anything goes", either.
Nor does "a nation of people is more than just replecable individuals" has "it's OK to exclude and persecute minority groups" as any kind of necessary consequence, anymore so than "this person doesn't belong to my family" means we can mistreat it.
It does however mean that "just bring foreigners in huge numbers to increase the local population" is not exactly a great solution. It might even amount to ethnic cleansing, and this has happened in the past too.
Historical immigration was often part of such power plays and land grabs (like against Aboriginals or Southern Africans), or "population engineering" (those in power bringing in populations to divide and conquer, and to replace and suppress local populations). In fact one of the common targets that was yielded against was exactly minority groups local to a nation or part of an empire.
It's only 'unsavory' if you're ignorant of the history and reality of Korea. Korea is a small, culturally and racially homogenous nation that, within living memory, was colonized by a power that attempted to wipe out its ethnic identity. It has no native racial minority groups. The ethnic and cultural unity of the nation is a source of pride for Koreans. You can't expect a nation like this to import millions of immigrants from other cultures.
The parent comment amounted to ethnically cleansing the Korean population. It might not involve guns and it might take a 3-4 generations but that is what happens when immigration doors are opened at the behest of corporation lobbied governments with no consideration for the local population.
The problem is that you’ve identified two distinct and non-overlapping sets of people as “everyone”. Everyone who was applauding 3.5 when it came out were industry hype people. Even the critical voices were industry hype people, paid to assume the AI is powerful and write about the possible negative consequences of that assumption.
Now we’ve all gotten familiar with 3.5, and we’ve come to understand its limitations, so the public knows it’s not a “godlike” AI.
Luckily there’s a fresh new model, not technically different from the earlier one but it cost more money to build. The hype group can start again, citing the publicly known limitations of 3.5. But in 6 months we’ll understand what’s wrong with it, and the public will be talking about the limitations, just in time for 4.5.
I recently used Jaccard similarity as a measurement of distance between two sets of online articles. It’s amazing how versatile it is for all sorts of weird tasks.
I uses to use Jaccard similarity combined with w-shingling at the character level to detect clusters of fraud sites. It was surprisingly effective, because it was able to pick up common patterns in the code even if they used completely different styles and text.
Interesting - I also used Jaccard similarity to classify clusters of malicious ad traffic schemes. The idea worked well. It was unclear if the similarity was due to mimicry or authorship, but that did not matter for our use.
