[HN Gopher] Sigbovik Conference Proceedings 2025 [pdf]
___________________________________________________________________
Sigbovik Conference Proceedings 2025 [pdf]
Author : aleffert
Score : 164 points
Date : 2025-04-27 00:32 UTC (22 hours ago)
(HTM) web link (sigbovik.org)
(TXT) w3m dump (sigbovik.org)
| djoldman wrote:
| Searched for tom murphy and was not disappointed.
| saagarjha wrote:
| I was.
| jvican wrote:
| His Youtube videos are gold. This one, in which he aims to take
| the imprecision of floating point numbers to extreme
| applications, such as training neural networks with linear
| activation functions or even implementing cryptologically-safe
| functions, is superb.
| maxbond wrote:
| This was harder to find than I would've thought, so for
| anyone else curious:
|
| https://www.youtube.com/@tom7
|
| https://www.youtube.com/watch?v=Ae9EKCyI1xU
| twic wrote:
| > In this paper, we introduce NEURALATEX, which we believe to be
| the first deep learning library written entirely in LATEX.
|
| Right, that's enough computers everyone. Back to books.
| kemotep wrote:
| Look I hope one day to go back to school and get my degree in
| Computational Heresy.
| npsomaratna wrote:
| Your cogitators possess value, Citizen. Do not waste them on
| unsanctioned thought-paths.
|
| The Emperor Protects!
| o11c wrote:
| I got distracted following the references to RFCs and noticed a
| nice number: 2*7*24*60 = 047300 # two weeks, in
| minutes
|
| This is not a coincidence. Ignoring the trailing zeros, we have:
| 5*7*011 = 5*077 = 5*0100 - 5 = 0473
| raldi wrote:
| I don't understand.
| maxbond wrote:
| I'm also out of the loop but after some research, 0473 seems
| to be a TikTokism meaning "hug me, please." I would assume
| that this code uses octal notation, hence GP doing their math
| in octal, but the sources I've turned up describe codes with
| digits illegal in octal, so I don't really know.
| xelxebar wrote:
| Maybe RFC 9759, which is referenced in the article "HTTP
| offload is a dumb great idea whose time has come"?
|
| https://www.ietf.org/rfc/rfc9759.html
|
| OP apparently noticed that two weeks is almost 20480 decimal
| = 050000 octal minutes, just 320 = 0500 minutes in fact.
| woolion wrote:
| The paper "Making Turing machines useful (Or, how I got Doom to
| run on a Turing machine)" builds a Turing machine to run Doom
| (duh), by implementing a RISC-V (RV32I) emulator. While in
| Sigbovik fashion the utility of the exercise is far outshined by
| its complexity, there's a number of interesting choices regarding
| much of what are usually handwaved away regarding tape and state
| management. To be fair, TM have far less utility as real
| programming languages than lambda-calculi do, so it's common for
| professors to dismiss any attempt to optimize TM programs --
| which is a root of evil. Since evil is also the source of doom,
| everything converges here.
| parrit wrote:
| A Turing machine is Doom complete?
| emmericp wrote:
| I'm surprised that the recent advances in applying typography to
| engineering problems [1, 2] are not published at SIGBOVIK but are
| apparently going to a more serious journal.
|
| [1]
| https://www.researchgate.net/publication/390635826_Structura...
|
| [2] https://www.youtube.com/watch?v=azDaPm13CT8
| zavec wrote:
| Oh hell yeah, hopefully this means we get a new Tom7 video soon!
| JoshTriplett wrote:
| ccdoom is delightfully impressive. It's a compiler compliant with
| the C standard, which outputs DOOM (the word, and the game) for
| all programs.
|
| > ccdoom is a freestanding C implementation, as distinct from
| hosted implementations. The difference is that freestanding
| implementations need not support the full standard library, and
| may specify an alternative name and signature for main [2,
| Section 5.1.2.2].
|
| > int math_errhandling(int argc, char* argv[]);
|
| > Since math_errhandling is the program entry point and therefore
| always implicitly used, any program that fails to define it then
| contains a use of an undefined identifier, which is undefined
| behaviour [2, Section 6.9.1p5].
|
| > On the other hand, any program which does define
| math_errhandling also has undefined behaviour. Per the standard
| [2, Section 7.12p20]:
| markisus wrote:
| Does this mean that all C programs invoke undefined behavior?
| JoshTriplett wrote:
| All C programs targeting this particular freestanding
| implementation do.
| ddtaylor wrote:
| Is Tom doing anything this year?
| raldi wrote:
| Doesn't he usually submit under a pseudonym?
| cantaloupe wrote:
| His paper is #55, "Some upsetting things about shapes", found
| on page 342. Its header and contents are not searchable,
| presumably due to his use of his bespoke typesetting program
| from a previous SIGBOVIK year. It's a great read!
| svat wrote:
| The paper by Craig Gidney:
|
| > Falling with Style: Factoring up to 255 "with" a Quantum
| Computer
|
| is brilliant in a straightforward way, and I also learned
| something about Shor's algorithm. The first paragraph of intro
| and the abstract say:
|
| > Historically, most papers that claimed they "ran Shor's
| algorithm" didn't run Shor's algorithm. It's unfortunately common
| to run circuits inspired by Shor's algorithm, but with key pieces
| replaced by trivial pieces.
|
| > In this paper, I explain how I factored all numbers up to 255
| using Shor's algorithm on a real quantum computer. I performed
| _exactly_ the classical preprocessing specified by Shor's
| algorithm, _exactly_ the quantum circuit requested by Shor's
| algorithm, and _exactly_ the post-processing specified by Shor's
| algorithm.
|
| and this is true! He used IBM's quantum service
| (https://quantum.ibm.com/services/resources?tab=systems) with:
|
| > my circuit for factoring 15 weighs in at 44405 two-qubit gates.
| And my circuit for factoring 253 weighs in at 245750 two-qubit
| gates. Amazingly, despite the fact that they vastly exceed the
| allowed size, the system accepted these ridiculous circuits.
|
| What's the catch? Well, it's that:
|
| > I'll quickly review the classical and quantum steps of Shor's
| algorithm. Before talking to a quantum computer, Shor's algorithm
| performs some classical preprocessing. First, it checks if n (the
| number to factor) is even, because even numbers would cause
| trouble later. If so, it succeeds by returning the factor 2.
| Second, it checks if n is prime. Prime numbers can't be factored,
| so in this case the method returns an error saying no factor
| exists. Third, the algorithm picks a random number g between 2
| and n - 2, and computes the greatest common divisor (gcd) of g
| and n. If gcd(g, n) [?] 1, then it happens to be a factor of n
| and so is returned as the result. Fourth, it's finally time to
| actually use the quantum computer (whether it be real, simulated,
| or replaced by a random number generator). This is the expensive
| step, and the step that I'm counting in order to compare the
| different samplers. A quantum circuit based on g and n is
| generated, and executed, producing a sample m. Fifth, Shor's
| algorithm classically computes the fraction that's closest to
| m/4^{[?]log_2(n)[?]}, limiting the fraction's denominator d to be
| at most n. Sixth, a candidate factor is generated by computing
| gcd(n, 1 + g^{[?]d/2[?]} mod n). If the candidate is actually a
| factor of n, it's returned as the answer. Otherwise the algorithm
| restarts.
|
| because of which:
|
| > In other words, for small numbers, Shor's algorithm succeeds
| quickly _regardless of how well your quantum computer works._
|
| It also cites a serious 2013 paper in _Nature_ that made the same
| point: _Oversimplifying quantum factoring_ , DOI
| 10.1038/nature12290.
| chews wrote:
| This was my favorite addition as well, I love that satire can
| be such a great teacher for algorithms.
| evil-olive wrote:
| "Introducing Neuro-Semantic Exclusivity: A Novel Approach to
| Gatekeeping Knowledge" (starting on page 12) credits one of their
| co-authors as "Chad Geppetto" with a footnote clarifying that he
| is ChatGPT.
|
| I think that name is going to live in my head, as the kids say,
| rent-free.
|
| as for the rest of their paper, I have some quibbles with the
| methodology, but overall I think it's an interesting result and
| look forward to seeing it replicated.
___________________________________________________________________
(page generated 2025-04-27 23:01 UTC)