qnleigh

19 ms·

331.
by qnleigh 2y ago
Can you elaborate? I think this is a really interesting question. It comes up over and over again, but it often feels like the two sides of the debate talk past each other. What does the mental model of 'iterative search through latent
332.
by qnleigh 2y ago
I wish I understood how mathematicians conceptualize the notion of 'right' and 'wrong' axioms in light of Gödel. I would've thought that the incompleteness theorems would be taken as evidence against Platonism, but
333.
by qnleigh 2y ago
I wish it would. Perhaps it will with rising costs of living and housing.
334.
by qnleigh 2y ago
Highly recommend this lecture by a working mathematician shared above ( https://www.youtube.com/watch?v=vYCT7cw0ycw [54min]). It's very much grounded in history and experience, much more so than in speculation. I wrote
335.
by qnleigh 2y ago
I think there's a pretty good case to be made that LLMs paired with automated theorem provers will become a useful tool to working mathematicians in the next few years. Another thread here links to a lecture from a professor of mathema
336.
by qnleigh 2y ago
That was wonderful, thank you for linking it. For the benefit of anyone who doesn't have time to watch the whole thing, here are a few really nice quotes that convey some main points. "We might put the axioms into a reasoning appa
337.
by qnleigh 2y ago
> it’s like something out of a sci fi movie where humans encounter an alien species that thinks in a way that humans cannot even comprehend due to biological limitations. I've increasingly felt this since GPT2 wrote that news piece
338.
by qnleigh 2y ago
That's a good question. They run the system on a small scale and validate there. The assumption is that no new error mechanism magically switches on when the simulation gets large enough, but it is did there would be no way to know. Ho
339.
by qnleigh 2y ago
I like the notion, implied in the article, that AGI won't be verified by any single benchmark, but by our collective inability to come up with benchmarks that defeat some eventual AI system. This matches the cat-and-mouse game we'
340.
by qnleigh 2y ago
Horse accidents aren't a leading cause of death though. Seems like it'll depend on how thoroughly self-driving becomes a thing of the past on its own.
341.
by qnleigh 2y ago
Well really it would be "you get a discount for the percentage of your mileage driven with a autopilot enabled." You can pay less if you drive in a setting that saves the insurance company money (on average).
342.
by qnleigh 2y ago
From Wikipedia: "The Standard Model of electroweak interactions has all the necessary ingredients for successful baryogenesis, although these interactions have never been observed[11] and may be insufficient to explain the total baryon
343.
by qnleigh 2y ago
A simple classical example where this is true is a repetition code. If you represent 1 as 11...1 and 0 as 00...0, randomly flip some bits, and then recover the logical state with majority voting, the probability of a logical error occurring
344.
by qnleigh 2y ago
What experimental data would you need to see to change your mind? Either from Google or from another player?
345.
by qnleigh 2y ago
That's a great analogy, and I basically agree with it. But there would be some ancient, abacus-wielding mathematicians who would be impressed by this fast-but-useless computer. One might take it as a sign that once/if the computer
346.
by qnleigh 2y ago
Depends on what you mean by "this type of device." Strictly speaking there are many efficiently verifiable quantum algorithms (including Shor's algorithm, the one that breaks RSA). But if you mean "this particular device
347.
by qnleigh 2y ago
I would add one more to this, which I would argue is the main reason: (0) For a quantum algorithm/simulation to be classically verifiable, it needs additional structure; something that leads to a structured, verifiable output despite t
348.
by qnleigh 2y ago
Great question. The device is fully programable. Arbitrary one-qubit operations and arbitrary two-qubit operations between adjacent qubits can be performed. Theoretically these are 'universal for computation', meaning that a large
349.
by qnleigh 2y ago
But to emphasize one quote from the blog: "the new experiment doesn’t add anything new to this old debate" about multiverse interpretations vs. something else. An equally viable and perhaps conceptually simpler interpretation of t
350.
by qnleigh 2y ago
It's currently believed that quantum computers cannot break all forms of public key cryptography. Lattice based cryptography is a proposed replacement to RSA that would let us keeping buying things online no problem.
351.
by qnleigh 2y ago
Nice, thanks for linking that paper. I also did below. The authors argue (e.g. in the first comment here https://scottaaronson.blog/?p=8310#comments ) that by their definition, Google still only has a fraction of one logical
352.
by qnleigh 2y ago
The peer-reviewed version at Nature has more technical details about the processor itself.
353.
by qnleigh 2y ago
and the arXiv preprint, which isn't paywalled https://arxiv.org/html/2408.13687v1
354.
by qnleigh 2y ago
I wouldn't call it extraordinary, as this has been expected since the first quantum error correcting codes were worked out theoretically. But it is a strong claim, backed up with comparably strong evidence. Figure 1d of the paper shows
355.
by qnleigh 2y ago
He's already blogged about it a bit here https://scottaaronson.blog/?p=8310#comments and here https://scottaaronson.blog/?p=8329 though I bet he will have more to say now that the paper is officially o
356.
by qnleigh 2y ago
This is a viable interpretation of quantum mechanics, but currently there is no way to scientifically falsify or confirm any particular interpretation. The boundary between philosophy and science is fuzzy at times, but this question is soli
357.
by qnleigh 2y ago
You need to distinguish between "physical qubits" and "logical qubits." This paper creates a single "first-of-a-kind" logical qubit with about 100 physical qubits (using Surface Code quantum error correction).
358.
by qnleigh 2y ago
Read documentation? Oftentimes you'll get a much better understanding of something by going to the source than by reading quick how-to's and Stack Exchange. This certainly goes for Python and Git.
359.
by qnleigh 2y ago
Right no one has ever gotten into MIT but couldn't afford it. This has literally never happened.
360.
by qnleigh 2y ago
Two other theories that could explain why OpenAI's models do so well: 1. They generate chess games from chess engine self play and add that to the training data (similar to the already-stated theory about their training data). 2. They

More ›