I hereby dub the type 14 torpedo from the 1940's a quantum torpedo. I know you might ask how a quantum torpedo can be powered by steam? When fired the torpedo had a 50ish% chance to hit an enemy ship therefore its quantum!
What indeed I might believe that in near future AI could emulate supporting "expert" participants in online meetings where one company serves as a vendor to the other company. Imagine there is a meeting, and you wish to convince the other company to buy your services. To an online meeting you bring couple of AI voiced experts who really can automatically answer couple of technical questions, which are trained in support of ChatGPT parameters and online sources. This way you can leave an impression to your client that you have many "engineers" supporting you, and thus secure a new business deal based on that impression. Also, if there is an online meeting about some major issue, you can use AI experts in the call to create an impression that many engineers are working on solving this problem. Usually, on such meetings camera is not turned on for all the participants, so your "experts" can participate online only with audio, and almost no one would find that being weird. To exploit AI on meetings you do not need to have key participants being replaced by AI, since this is silly, and explained in the video, but as I mentioned, unfair advantage can be made by having those fake supporting participants. We can already find videos on RU-vid where AI avatars are emulated to respond in real time like real people, and this is not easy to notice even when you see the video of an AI generated person. Therefore, such technology already exists, and it would be interesting to see if someone starts utilizing these in order to trick other parties.
This whitepaper clearly explains how post-Quantum end-to-end encryption works: drive.google.com/file/d/1RyED7xJW6sBB6MBwDFuTHCn1dmdfgTP3/view?usp=sharing
Tech bros and finance bros operate in the same way: hoping to confuse you by sounding smart so you don't think about how actually it's dumb. But it mostly works, though, so. Why would they stop doing it?
Why clones in the first place? If all of this data exists and AI is this phenomenally good, why would we ever want to limit it to functioning like me or you or any other fallible human? Why would AI need to attend zoom meetings with other AI when they can all communicate instantaneously in binary code and access the entire internet of information. Like, what?
Angela you're a fantastic science communicator but the words "Eric Yuan CEO Of Zoom" (no comma) are now tattooed on the inside of my prefrontal cortex forever, and it's your fault.
The scale at 23:30 cracked me up. I took a screenshot of just the scale and put it in my college's computer science students' Discord meme channel. (and of course linked this video, where the scale originates from)
The whole part about quantum cryptography was wrong. Quantum computers are not just very fast computers, they don't break encryption by brute force, and using post-quantum cryptography for things that can be stored and decrypted in the future (meetings) and not for things that must be decrypted now (authentication) makes perfect sense.
I recommend episode 280 of Prof Sean Carroll's Mindscape podcast where he interviews François Chollet and he explains how large language models are not as intelligent as they can sound, because what they are doing is substituting massive amounts of memorisation for actual understanding.
Damn you are so smart Angela. You probably think “this is obvious, doesn’t everyone see this the way I do?” Sadly, no. No they don’t. Keep preaching sensible truths!!!
Unpopular opinion but digital ai twin isn't a dumb idea. Eric Youn the Ceo of Zoom did bad job of explaining it. The idea is every employee will have a personal knowledge base that others can ask dumb questions without having to schedule an entire meeting for said dumb question. You can technically create a knowledge base with google docs but nobodies gonna search through that to find the info needed. This were the "Gen Ai" llm rag comes in handy for basically approximate information retrieval from your knowledge base when someone needs to you a stupid question. You made a point about their be information in your head that the ai won't have ...the paradime shift is that employeers with just require you to create content build out your knowledge base
The claims Angela makes about post-quantum encryption are either wrong or overly simplyfied to the point of being unreliable. We absolutely want post-quantum encryption *before* a useful quantum computer is made. If it is implemented after quantum computers good enough to break current encryption exist, then it's already way too late. @3dcbd393ce 's comment goes more in depth. If someone is reading this comment and wants information from an expert I recommend searching for his comment.
Great video as always Angela! However, I do think that the read on the post-quantum encryption isn't quite right. Public key cryptography has two steps, the key exchange and the encryption/decryption itself. In the first step you have a form of authentication a key is passed to you. Then in the second stage you use the key to decrypt. Basically, you have one step where you have to give the right person the right key, and then a stage where they use the key. The way I read the white paper (and some of the associated algorithms) is that they use a cryptographic method which is quantum-secure, or in more concrete terms the decryption process cannot be brute forced by a quantum FFT and thus it's not in the class of problems which is hard for classical compute but easy for quantum compute. However, they do not apply a quantum-safe cryptographic method to the authentication and key exchange process. Or in the simple terms, they have quantum-safe locks, but the way they figure out who to pass the right keys to is not quantum-safe. In a way that's sensible because what it means is that if you recorded a video of your secret baking recipe with your grandmother or sharing secret conspiracy details about the Boston Molassacre (fellow MITer here, gotta share the secret tales of our neighbor), those recordings would be quantum cryptographically safe and thus if such computers ever existed (which I'm not bullish about, but if) then that video would be safe. The way it could be unsafe is if you used your quantum computer during the key exchange / authentication process to effectively crack the authentication and fake in such a way to get a key that you're not supposed to. But that could only happen if you had a quantum computer today when you do the authentication. So what it's basically saying is, recorded videos are safe and the only way this scheme is unsafe is if you had a quantum computer today to perform an attack to get the valuable keys, but since you don't have a quantum computer today you won't get those keys so we won't make that part quantum-secure. That's actually pretty sensible, or at least, it's a sensible way of doing something that probably isn't sensible. You can definitely argue that the quantum-safe encryption on a Zoom recording is overkill (and that would be easy to argue), but if you were going to do it, cutting this corner is probably a smart idea to make it at least cheaper to do, and you're safe as long as the NSA or Russia doesn't have a secret quantum computer which today is already capable of breaking the encryption. So don't get me wrong, a quantum-safe cryptographic method on Zoom videos is pretty silly, but I do think they are justified in the corners they cut there.
It doesn't even make much sense to use the stack as a metaphor, because the further a thing is down the stack, the older it is or the least priority it has. The only aspect of the stack that is relevant is that things at the bottom will be removed last, but might as well talk about a list. Unless you conceptualize the stack as a series of potential human discoveries stacked up by God, I guess. Or perhaps it's a way of saying that on our way to creating that technology, new things might stack up as we unstack it, because a certain discovery might lead to new explorations of greater priority or new finer issues to be solved in order to unstack further. I don't know. They probably just use that expression to sound tekky and altogether omit those technicalities. But it does sound funny.
I agree with most of what you say here but I think the comparison to quantum cryptography is not a good one. We have models of what quantum computers could do if they could be built at a reasonable scale; it is reasonable to try to defend against that in various ways assuming the models are at least somewhat accurate. (I have no opinion on how well Zoom's thing does that, though.)
William Keith a.k.a Ian Douglas and Polish writer Jack Dukaj have explored consequences of such AI stand-ins in Galaxy Corps/Starcarrier and "Black Oceans" books, respectively.
I didn't know this happened with Kaku, but I feel vindicated because I when heard him weigh in on some philosophical topic years prior to that news appearance he was absolutely dead to me...but I couldn't even remember what he said, and sometimes I wondered if I was being unfair. nope, he's got the same disease that many others seem to have caught, including Neil DeGrasse Tyson. how has Bill Nye avoided this trap? but also, Pinker and Gladwell are kinda equally terrible at getting science correct in general. there's a really good podcast episode about Gladwell on If Books Could Kill
Useful: Having an AI that I can ask to make a reservation at some restaurant, and that restaurant having an AI that can accept that reservation, so that there are AI interfaces that AI easily can integrate with. I don't care if it's through natural language or whatever. I even can imagine a not-bad version of this being that I can ask my AI assistant to clarify with your AI assistant "what did you mean with that thing you said", where AI assitant asks you to clarify, and that assistant talks to my assistant that then let's me know. It might be a new asynchronous (because I will not be interrupted by my AI assistant) way to clarify stuff that email is tremendously bad at doing. But why should my AI assistant ever be a clone of me?
If this AI clone thing actually worked as it "should", meaning the clones actually acted like you in a meeting, the the clones would learn things you don't know - at least if it's an actual useful meeting - and won't learn because you weren't in the meeting, making you superfluous in comparison to your clone. The whole thing to me reads as "Eric Youn, the CEO of Zoom's job is going to useless meetings where nothing actually happens and read useless emails all day, and he doesn't even notice."
You did read all the comments, you said, right? So maybe you might see this one, too, which I think would be intriguing and scientifically useful to you: Inflation of space doesn't exist, and what instead is causing the preliminarily invented cosmological red-shift phenomenon is instead gravitational red-shift of light, because the initially by generation III stars produced cold dark matter neutrinos long-term stay confined within comparatively small regions, namely galaxies rather than spreading out to inter-galactic space, allowing annihilation of this cold dark matter mass (in likely more or less exponential decay manner over time), which means that from the further away - and hence from older times when the cosmological scale intergalactic gravitational well was still deeper - that a photon comes into any remote region that's billions of years into the future where far less cold dark matter is remaining, it's climbing up a cosmological gravitational well and is red-shifted that way. This should also resolve the horizon problem and the crazy super-luminal space expansion speculation, rule out tired light, and imply the CMB was even more energetic in the past.
I feel like, when it comes to what is and isn't illegal for people to promise, the big tech cult has been given practical immunity from any consequences for falsely promising results
I have been unimpressed with the savage glee Tech Bros express at the prospect at leaving entire categories of the workforce, like say, the entire artistic community, without a way to earn a living , beggars in the street. They have taken the possibility of AI automation as a knife in the worker's collective neck, demanding obeisance and to work for a pittance, least your job is on the chopping block. So no, I neither going to support nor accept anything they have to offer since they're clearly my class enemies and do not have my best interest at heart.
5:00 - Dear Angela .... if people in USA were questioning numbers, we wouldn't get movies like Margin Call or Big Short, Elon Musk would be drugged mumbling junkie, Elizabeth Holmes would be waitress in McDonald, WeWork and Uber wouldn't be called 'tech companies' .... you get the idea? ;)
