Shtetl-Optimized

The best discussion I've seen so far is from Scott Aaronson: https://scottaaronson.blog/?p=6823 > To illustrate, in the special case that GPT had a bunch of possible tokens that it judged equally probable, you could simply choose whichever token maximized g [a cryptographic function]. The choice would look uniformly random to someone who didn’t know the key, but someone who did know the key could later sum g over all n-grams and see that it was anomalously large. The general case, where the token probabilities can all be different, is a little more technical, but the basic idea is similar.

Every single discussion thread about any announcement in quantum feels exactly the same: 1. this is not what you think it is, there are caveats and the title is misleading. 2. I feel there is so much hype around quantum computing. 3. It's like nuclear fusion, always 10 years away. 4. Nuclear fusion is at least practical. QC is not. 5. Did they manage to find prime factorization of 17 already? LOL 6. Cryptography will be dead, RIP bitcoin and https. 7. No, actually quantum computers are useless. It's tiring. I recommend this recent post by Scott Aaronson https://scottaaronson.blog/?p=9668

Scott Aaronson wrote a bit about the following thought [0]. If copying a brain and simulating reality ala The Matrix is possible at all, then if you get your brain copied you live one biological live but your copies have an unbounded number of existences (millions? billions? trillions?) So, if copying brains is possible, and you don't know which version of you you are, you might have odds of, say, 1 to 1 trillion to be living your first, biological live. Which is to say, if copying brains is possible, you are likely to be running in a simulation already. [0] there's multiple links and I can't find where I first read, but I found this one from 2024, https://scottaaronson.blog/?p=7774 and uhh.. turns out the argument isn't from him personally (and he doesn't even believe on it), and is best presented here https://simulation-argument.com/ (though it's presented very differently so idk)

I can't entirely follow the details, but apparently quantum mechanics actually doesn't work for fields other than C, including quaternions. https://scottaaronson.blog/?p=4021

This article was on the front page recently: https://scottaaronson.blog/?p=9534 So at least some lay people easily realized he wasn't worth getting involved with.

Isn't it rather that the Busy Beaver function is uncomputable, particular values can be calculated - although anything great than BB(5) is quite a challenge! https://scottaaronson.blog/?p=8972

Yeah, but the vexing part is "how can that be true for e.g. N=643 but not N=642"? What happens at whatever number it starts true at? Incidentally, Gödel's theorem eventually comes down to a halting-like argument as well (well, a diagonal argument). There is a presentation of it that is in like less than one page in terms of the halting problem---all of the Gödel-numbering stuff is essentially an antiquated proof. I remember seeing this in a great paper which I can't find now, but it's also mentioned as an aside in this blog post: https://scottaaronson.blog/?p=710 wait jk I found it: https://arxiv.org/abs/1909.04569

My favorite read on the subject: https://scottaaronson.blog/?p=2756

It's not clear how your perception of quantum computing would lead to 'faster horses' in the current view of NN architectures - keep mind that the common view of 'exploring many paths simultaneously' is at best an oversimplification (https://scottaaronson.blog/?p=2026). That said, perhaps advances in computing fundamentals would lead to something entirely new (and not at all horselike).

Thanks, but you can follow the comments on the answers to get eg https://scottaaronson.blog/?p=2196 See https://scottaaronson.blog/?p=8329 for something more recent.

Aaronson wrote a blog entry today about the interview: https://scottaaronson.blog/?p=8200 (He recommends viewing the interview at 2x speed.)

https://scottaaronson.blog/?p=8088 > You need to find a mathematical reason why it can’t halt, and there’s no systematic method for finding such reasons—that was the great discovery of Gödel and Turing nearly a century ago. That's only true for sufficiently large N, large enough to encode the halting paradox. (How large is that N? It's larger than 5!) Smaller N can and do fall to systematic methods. > (x) = (5x+18)/3 if x = 0 (mod 3), The second = should be \equivalent. This is a rare case where that actually matters, because is being used in both non-modular integer operations (divide by 3) and modular operations (where division by 3 is not defined).

Not me :) unequal outcomes is either plain observation, or fantasy --- only, the latter only, I might deign to call "beliefs".. Given, some may put observation and fantasy in the same equivalence class, I should like to encounter them. (Talented SWEs?/intermediate reppers(aka designers)?) EDIT: in case you missed it, follow-up on ANK (not KANs) https://scottaaronson.blog/?p=762

See https://www.scottaaronson.com/papers/optics.pdf “We give new evidence that quantum computers—moreover, rudimentary quantum computers built entirely out of linear-optical elements—cannot be efficiently simulated by classical comput- ers. In particular, we define a model of computation in which identical photons are generated, sent through a linear-optical network, then nonadaptively measured to count the number of photons in each mode. This model is not known or believed to be universal for quantum com- putation, and indeed, we discuss the prospects for realizing the model using current technology. On the other hand, we prove that the model is able to solve sampling problems and search problems that are classically intractable under plausible assumptions.” Which is the basis for the experiment discussed here: https://scottaaronson.blog/?p=5122

I'm not even sure that a quantum computer has incontrovertibly outperformed a classical one at this time on anything. Some have claimed it, on problems that have no practical utility. Classical results have also been improved on response. Arguably then the only concrete achievement of quantum computing to date has been improvements to classical algorithms. However, I fully expect that we will eventually have powerful quantum computers. Practical applications for those are things like modelling chemical reactions or quantum systems, or optimisation. No discussion on quantum computing would be complete without a link to Scott Aaronson's blog. Worth a read to get a better understanding of the subject. https://scottaaronson.com/

Yes, here's Scott Aaronson's opinion on that measure: https://scottaaronson.blog/?p=4649 TL;DR: Instead of using a single measure, just give all the details on the machine as clear as possible: > How many qubits do you have? With what coherence times? With what connectivity? What are the 1- and 2-qubit gate fidelities? What depth of circuit can you do? What resources do the standard classical algorithms need to simulate your system? Most importantly: what’s the main drawback of your system, the spec that’s the worst, the one you most need to improve? What prevents you from having a scalable quantum computer right now? And are you going to tell me, or will you make me scour Appendix III.B in your paper, or worse yet, ask one of your competitors?

There's a lot of introduction here, but it starts getting to the point in section 3.4 "The previous paper". That paper is "A Relatively Small Turing Machine Whose Behavior Is Independent of Set Theory" https://arxiv.org/abs/1605.04343 (and an improvement in the blog comments on https://scottaaronson.blog/?p=2725) I recommend starting there just to know where the conversation is going and then coming back to this.

Here’s a recent quote from Scott Aaronson: > In truth, anything that a QC can calculate, a classical computer can calculate as well, given exponentially more time: for example, by representing the entire wavefunction, all 2n amplitudes, to whatever accuracy is needed. That’s why it was understood from the very beginning that quantum computers can’t change what’s computable, but only how efficiently things can be computed. https://scottaaronson.blog/?p=7321

He writes a great reply to this in his own comment section: "Michael #3: In a certain sense you’re right. The language models now being adopted by millions of programmers don’t write working code; they only seem-to-write-working-code. They’re not, unfortunately, already doing millions of students’ homework for them; they’re only seeming-to-do-the-homework. Even if in a few years they help me and my colleagues do our research, they won’t actually be helping, but only seeming-to-help. They won’t change civilization; they’ll only seem-to-change-it." [0] https://scottaaronson.blog/?p=7094#comment-1947189

In some contexts I agree, like when field experts publish blogs or tweets. For example, Scott Aaronson does this often ( https://scottaaronson.blog/ ) The problem is that this generally requires already having the credibility / “clout”. Without that, you’re just some anonymous person screaming into the void that the mainstream is wrong. there isn’t a dialogue, just a rant. And without a name backing you up, the internet has little reason to care about the rant. I don’t see the numbers and analysis in this article contributing to any discussion but rather providing a facade of authenticity for the following political rant. And the problem is, if someone says it’s legit analysis, why should I believe them? Until an expert will put their name on the line, I just don’t care; and that brings us back to the peer review / “clout” problem

In case anyone is missing the reference the title is based on: https://en.wikipedia.org/wiki/Champagne_for_my_real_friends,_real_pain… Also, here is a great takedown by Scott Aaronson: https://scottaaronson.blog/?p=6871

They should go back to the drawing board and actually get a stronger argument they really have a quantum computer. For those not following https://scottaaronson.blog/?p=6645, their quantum supremacy experiments used quite a bad metric that's very easily spoofable. https://arxiv.org/abs/2112.01657 Quantum computer is the perpetual motion machine of the 21st century. I'm sure we'll never get there, but getting an insight like the second law of thermodynamics of why it's impossible will be important too. And I don't buy the attitude of "the errors will be bad", that's like "there is always friction" to perpetual motion machines. There needs to be something deeper.

I think you are having a relatively common misunderstanding. Indeed, just having some analog process run itself is a boring example of "supremacy". https://www.smbc-comics.com/comic/2013-07-19 A "useful" computation is indeed the important final proof of supremacy. But intermediary results of "programmable" but useless computation is a pretty amazing achievement on its own, and it is not just "an analog process running itself". The distinction is described quite well here and it really matters to computational complexity theory https://scottaaronson.blog/?p=5460

A great writeup about Shor's algorithm linked in the article: https://scottaaronson.blog/?p=208

Are quantum computers being used in economically significant ways that only are possible with quantum computers yet? If so, what is the best example? Last time I checked few years ago, my understanding was the answer was still no; specifically, believe I asked Scott Aaronson as a follow up question to a talk he gave: https://www.scottaaronson.com/

Also copied from a couple blog posts ago he doesn't self-identify as a physicist either. https://scottaaronson.blog/?p=6457 I also had the following exchange at my birthday dinner: Physicist: So I don’t get this, Scott. Are you a physicist who studied computer science, or a computer scientist who studied physics? Me: I’m a computer scientist who studied computer science. Physicist: But then you… Me: Yeah, at some point I learned what a boson was, in order to invent BosonSampling. Physicist: And your courses in physics… Me: They ended at thermodynamics. I couldn’t handle PDEs. Physicist: What are the units of h-bar? Me: Uhh, well, it’s a conversion factor between energy and time. (*) Physicist: Good. What’s the radius of the hydrogen atom? Me: Uhh … not sure … maybe something like 10-15 meters? Physicist: OK fine, he’s not one of us.

This should have a (2005) --- I was very surprised to see that the competition deadline was 17 years in the past! In any case, here's the follow-up with the winners: https://scottaaronson.blog/?p=42

If you want to understand math behind Quantum Mechanic and Quantum Computing I think this is a great resource: https://scottaaronson.blog/?p=4805

"A key sticking point in the approval process has been the framework’s recommendation that teachers refrain from labeling students as “naturally talented” in math. This has led to accusations from parents and educators that it holds back “gifted” students." Along with NYC mayor deBlasio seeking to terminate that city's gifted and talented education programs as well as Virginia's recent approach to "equity" by decontenting gifted education. The examples demonstrate the illiberal trends we are seeing in the U.S. wherein novel concepts like "equity" replace "equality" and merit is seen as racist. See https://www.nytimes.com/2021/10/08/world/nyc-schools-will-replace-the-… See also https://scottaaronson.blog/?p=4979 See also https://bariweiss.substack.com/p/mit-abandons-its-mission-and-me

> If you claim LLMs produce thought, it's equivalent to claiming light bulbs undergo fission You're making a huge logical leap. There is nothing equivalent about these claims other than that they are made in English. > calling a language producing machine necessarily a thinking machine is an old old mistake. Nobody claims that all language models think. Small markov chain language models clearly aren't thinking and produce a lot of gibberish. The difference is that, to the surprise of many people several years ago, but to the surprise of nobody who has been following along today, the corpus of all text produced by humans contains within it information about how the world works and also information about how to reason. Using that corpus to train a sufficiently large language model causes the language model to learn a world model and a reasoning model, which can be used to perform longer chain thinking with test time compute techniques. People who think deeply for a living recognize thinking when they see it. https://scottaaronson.blog/?p=9979