Hacker News: kccqzy

New comment by kccqzy in "CrankGPT"

kccqzy — Mon, 15 Jun 2026 15:43:22 +0000

140 watts is the FTP. That means you can do it for an hour, and it will be an extremely exhausting hour and you will want at least two days of rest to recover from this workout before doing it again.

If you are not chasing watts, it’s much more sustainable to do 70 watts for two hours. You can probably do this every day.

New comment by kccqzy in "CrankGPT"

kccqzy — Mon, 15 Jun 2026 15:38:55 +0000

I just rode with an untrained cyclist (new to cycling) yesterday. The person averaged 80W over five hours. It’s about right for an actual untrained cyclist.

New comment by kccqzy in "CrankGPT"

kccqzy — Mon, 15 Jun 2026 15:33:31 +0000

It’s basically well known to cyclists that training with a power meter that tells you “watts” more accurately gauges effort and caloric expenditure. (Heart rate gauges subjective effort however, taking into account stress, caffeine intake, etc.)

It’s also interesting that the industry has settled on using watts to mean rate of useful work whereas calories to mean the total work including inefficiencies, despite that calories is just a unit of energy. A rule of thumb for cyclists is that in addition to usual unit conversions, the “calories” figure should be multiplied by four to account for energy expended by the body but not used for rotating the pedals. I don’t use rowing machines but I’m sure they would have a similar conversion factor in order to calculate calories.

New comment by kccqzy in "Noise infusion banned from statistical products published by Census Bureau"

kccqzy — Sun, 14 Jun 2026 01:02:04 +0000

An earlier article on the same blog had some very useful information on how easily the aggregated census data from 2010 (before differential privacy) could be used to reconstruct real data for individuals: https://desfontain.es/blog/us-census-reconstruction-attack.h...

I am personally convinced that the reason noise infusion was banned was because powerful people were already reconstructing individual data from census for the purpose of gerrymandering, and they wanted to continue gerrymandering.

New comment by kccqzy in "Statement on US government directive to suspend access to Fable 5 and Mythos 5"

kccqzy — Sat, 13 Jun 2026 05:09:38 +0000

Marketing is never ever to blame. Remember a few months ago when the U.S. government labelled them a supply chain risk? What eventually happened was that a federal judge issued a temporary injunction while calling it a "classic First Amendment retaliation." The Constitution protects such marketing; the government is not allowed to be maddened by such marketing.

New comment by kccqzy in "Statement on US government directive to suspend access to Fable 5 and Mythos 5"

kccqzy — Sat, 13 Jun 2026 02:11:34 +0000

Realpolitik in action. Great powers just impose export controls because they know they can and they think it would be beneficial to the nation.

New comment by kccqzy in "Swift at Apple: Migrating the TrueType hinting interpreter"

kccqzy — Sat, 13 Jun 2026 01:42:45 +0000

Hinting purposefully destroys letter shapes in exchange for crispness. People who like macOS style font rendering prioritize letter shapes faithful to the font designer more than crispness.

Whether good letter shapes is more legible or crisper text is more legible is basically subjective. In the 2000s before HiDPI became popular different people really thought one was more legible than the other and vice versa. HiDPI made this basically moot.

New comment by kccqzy in "AI agent bankrupted their operator while trying to scan DN42"

kccqzy — Fri, 12 Jun 2026 20:50:23 +0000

It’s just a consensus that’s implicit and unstated.

New comment by kccqzy in "A PDF that changes based on how its read"

kccqzy — Fri, 12 Jun 2026 19:26:50 +0000

Why would you use the exact same technique? Remove all fonts and all text from the PDF and render everything as vector graphics. It’s an old trick to prevent people from extracting paid commercial fonts from your PDF.

And of course, OCR doesn’t work here just like it doesn’t work for the original use case.

New comment by kccqzy in "Adaptive PDFs"

kccqzy — Fri, 12 Jun 2026 19:18:35 +0000

The author says LaTeX doesn’t produce tagged PDFs; but that’s entirely because most users of LaTeX didn’t care enough. All the pieces are there. We just need more user education.

New comment by kccqzy in "How Terry Tao became an evangelist for AI in math"

kccqzy — Fri, 12 Jun 2026 17:17:33 +0000

You can look forward to that, but today I’m already experiencing something worse but close enough for all but the most critical code: AI-driven algorithm design + tests in your favorite property-based testing library (like OG QuickCheck in Haskell or hypothesis in Python).

Of course problems remain in both approaches: a human or AI needs to make sure the lean proof is proving the correct translation from natural language spec to a formal theorem, or the PBT is testing the right properties translated from natural language.

New comment by kccqzy in "Software is made between commits"

kccqzy — Fri, 12 Jun 2026 03:39:39 +0000

That's not a jj feature, but a CitC feature. You can use it on hg workspaces or plain p4 workspaces.

New comment by kccqzy in "Faking keyword arguments to functions in C++"

kccqzy — Fri, 12 Jun 2026 03:05:38 +0000

That's a C99 feature, designated initializer. Hardly modern. Yes it was ported to C++ relatively late, but it happened in C++20.

New comment by kccqzy in "Anthropic requires 30 day data retention for Fable and Mythos"

kccqzy — Wed, 10 Jun 2026 22:43:41 +0000

That’s exactly what my employer had communicated. It will not be allowed.

New comment by kccqzy in "Anthropic's model naming, extrapolated"

kccqzy — Wed, 10 Jun 2026 20:06:16 +0000

Claude Littera: a model so good at summarizing that it summarizes everything to a letter.

(Unfortunately just like English a letter can also mean a correspondence.)

New comment by kccqzy in "Building an HTML-first site doubled our users overnight"

kccqzy — Wed, 10 Jun 2026 15:48:46 +0000

My biggest tip to reduce complexity of data validation if you are using React is to stop using React controlled components and switch to React uncontrolled components. They are an underused part of React. You usually don’t need React to handle every single keypress and every single character being entered by the user. In fact before React popularized it, it was unusual for form components to update on each key press; traditional desktop apps tend to validate when a field loses focus only, not on each key press. This has at least three benefits: (a) good for performance, (b) reduces unnecessary error UI when the user is in the middle of entering data, and (c) simplifies your own code by not having to deal with prefixes of valid input that’s not itself valid.

New comment by kccqzy in "All 9,300 Japanese train station, animated by the year it opened (1872–2026)"

kccqzy — Wed, 10 Jun 2026 14:48:23 +0000

I hate the History API especially pushState. Even with this limit of 100 times per 10 seconds it still pollutes my browsing history too much. I need to vibe code an extension that makes pushState/replaceState noops on all webpages.

New comment by kccqzy in "xAI is looking more like a datacentre REIT than a frontier lab"

kccqzy — Mon, 08 Jun 2026 20:06:17 +0000

The $80B of equity raise is nothing compared to its previous share buyback programs.

New comment by kccqzy in "Stop the Apple Music app from launching"

kccqzy — Mon, 08 Jun 2026 18:36:24 +0000

Ah a long time ago when the Music app was still called iTunes I have configured all music files to be opened with quicktime player. It’s been so long that I forgot the default was the Music app. To me it’s absolutely clear that playing a file doesn’t mean I want it in my library.

New comment by kccqzy in "Do we fear the serializable isolation level more than we fear subtle bugs (2024)"

kccqzy — Mon, 08 Jun 2026 17:14:04 +0000

It’s somewhat rare for a transaction to read many rows and then write. Typically a transaction reads a single row, applies business logic, writes a single row. Then you have analytic workloads that do no writes and read entire tables.