Hacker News: sanxiyn

New comment by sanxiyn in "A Survey of Inlining Heuristics"

sanxiyn — Tue, 09 Jun 2026 05:57:31 +0000

I know it is old, but the classic is a classic for reasons. It is criminal this survey does not include the venerable Secrets of the Glasgow Haskell Compiler inliner by Simon Peyton Jones et al.

New comment by sanxiyn in "Constraint Decay: The Fragility of LLM Agents in Back End Code Generation"

sanxiyn — Mon, 25 May 2026 08:50:50 +0000

GPT-5.2 was released after November 2025.

New comment by sanxiyn in "Samsung chip workers will get an average $340k bonus as AI profits soar"

sanxiyn — Fri, 22 May 2026 04:25:23 +0000

Consumers have a choice of not buying, so generally they do not lose to unions.

New comment by sanxiyn in "Samsung chip workers will get an average $340k bonus as AI profits soar"

sanxiyn — Fri, 22 May 2026 02:59:42 +0000

The union makes us strong.

-- A proud programming union worker in South Korea since 2018.

New comment by sanxiyn in "OpenBSD 7.9"

sanxiyn — Tue, 19 May 2026 14:18:56 +0000

"Parking" lock is a reference to this:

https://webkit.org/blog/6161/locking-in-webkit/

New comment by sanxiyn in "Project Glasswing: what Mythos showed us"

sanxiyn — Tue, 19 May 2026 03:12:25 +0000

Mozilla published some numbers and actual bugs.

https://hacks.mozilla.org/2026/05/behind-the-scenes-hardenin...

New comment by sanxiyn in "A few words on DS4"

sanxiyn — Fri, 15 May 2026 01:28:00 +0000

There is a benchmark for performance work, and I think it is not being optimized by model vendors. The latest result from GSO is that both Opus 4.6 and 4.7 slightly outperforms GPT 5.5. This also matches my experience.

https://gso-bench.github.io/

Aurora: A Leverage-Aware Optimizer for Rectangular Matrices

sanxiyn — Sat, 09 May 2026 05:02:36 +0000

Article URL: https://blog.tilderesearch.com/blog/aurora

Comments URL: https://news.ycombinator.com/item?id=48071970

Points: 1

# Comments: 0

Spec CPU 2026

sanxiyn — Thu, 07 May 2026 02:36:52 +0000

Article URL: https://www.spec.org/cpu2026/

Comments URL: https://news.ycombinator.com/item?id=48044759

Points: 2

# Comments: 0

New comment by sanxiyn in "Will you heed my warnings now?"

sanxiyn — Thu, 30 Apr 2026 11:00:27 +0000

No you really can't. Being able to factor 15 but not 21 with Shor's algorithm is normal. I know it sounds absurd, but it really is that way. Because factoring 21 is about 100x times harder than factoring 15.

See https://algassert.com/post/2500 for details.

New comment by sanxiyn in "Will you heed my warnings now?"

sanxiyn — Thu, 30 Apr 2026 09:02:42 +0000

The main point is that just as you can't ask for tiny nuclear explosion because nuclear physics just doesn't work that way, you also can't ask for factoring of 21 with Shor's algorithm. Quantum computing just doesn't work that way, sorry.

New comment by sanxiyn in "Will you heed my warnings now?"

sanxiyn — Thu, 30 Apr 2026 07:54:25 +0000

I will let Scott Aaronson speak. (See https://scottaaronson.blog/?p=9668)

> Sometimes these days, I'll survey the spectacular recent progress in fault-tolerance, 2-qubit gate fidelities, programmable hundred-qubit systems, etc., only to be answered with a sneer: "What's the biggest number that Shor's algorithm has factored? Still 15 after all these years? Haha, apparently the emperor has no clothes!" I've commented that this is sort of like dismissing the Manhattan Project as hopelessly stalled in 1944, on the ground that so far it hasn't produced even a tiny nuclear explosion... If there's a reason why you think it can't work beyond a certain scale, say so. But don't fixate on one external benchmark and ignore everything happening under the hood, if the experts are telling you that under the hood is where all the action now is, and your preferred benchmark is only relevant later.

New comment by sanxiyn in "There Will Be a Scientific Theory of Deep Learning"

sanxiyn — Sat, 25 Apr 2026 08:57:27 +0000

As I understand, Rice's theorem does not apply because neural networks are not Turing-complete.

New comment by sanxiyn in "Mozilla Used Anthropic's Mythos to Find and Fix 271 Bugs in Firefox"

sanxiyn — Wed, 22 Apr 2026 04:39:28 +0000

To some extent yes, but models are good at reverse engineering such that it isn't as great advantage as you might think.

New comment by sanxiyn in "Mozilla Used Anthropic's Mythos to Find and Fix 271 Bugs in Firefox"

sanxiyn — Wed, 22 Apr 2026 04:36:06 +0000

Last three CVEs are collections of bugs. CVE-2026-6784 is a collection of 55 bugs. CVE-2026-6785 is a collection of 154 bugs. CVE-2026-6786 is a collection of 107 bugs.

As for credits, I think bugs are ultimately credited to people, and this time Mozilla people used Mythos, as opposed to Anthropic people using Opus or Mythos.

New comment by sanxiyn in "My practitioner view of program analysis"

sanxiyn — Tue, 21 Apr 2026 23:12:13 +0000

Why do you think so? I didn't get such impression.

New comment by sanxiyn in "My practitioner view of program analysis"

sanxiyn — Tue, 21 Apr 2026 22:56:50 +0000

The gap between formal and informal has been pointed out as an Achilles' heel of formal methods from the dawn of the field, so critique is not particularly new. The standard reference is Social processes and proofs of theorems and programs (1979), which is worth reading.

Measuring Human Performance on ARC-AGI-3

sanxiyn — Tue, 14 Apr 2026 23:12:35 +0000

Article URL: https://arcprize.org/blog/arc-agi-3-human-dataset

Comments URL: https://news.ycombinator.com/item?id=47772654

Points: 2

# Comments: 0

New comment by sanxiyn in "Cloudflare's Gen 13 servers: trading cache for cores for 2x performance"

sanxiyn — Fri, 27 Mar 2026 09:06:39 +0000

Well, that's because this post is about Gen 13. In FL2 post (presumably on same Gen 12 servers), they say 25% lower latency.

New comment by sanxiyn in "Day 1 of ARC-AGI-3"

sanxiyn — Fri, 27 Mar 2026 03:37:54 +0000

In this case the code is public and you can see they are not cheating in that sense.