Hacker News Reader

queueberton Show HN: Flash Attention in ~100 lines of CUDA

As a person who finds CUDA extremely easy to write and integrate, what does Triton have to offer?

queueberton Reversing for dummies – x86 assembly and C code (Beginner/ADHD friendly)

I had the same thought. Couple it with a source fuzzer to generate a large training dataset of valid source, and you could make a very nice disassembler.

queueberton NASA engineers make progress toward understanding Voyager 1 issue

Or maybe hardware overkill. I have a soft spot for small, dedicated computers like the Apollo Guidance Computer that have physical buttons and simple functionality. The DDIs on jet fighters are another example.

queueberton Glassdoor updated my profile to add my real name and location

I find it hilarious that all the money pumped into Glassdoor has created less useful information about companies than the Better Business Bureau.

queueberton Is Cosine-Similarity of Embeddings Really About Similarity?

Yes, this would be fast assuming the vectors are pre-normalized, obviating the need to repeatedly calculate two square roots.

queueberton Pilot of Boeing flight says he lost control after instrument failure

On a related note, one of the whistleblowers was just found dead: https://www.msnbc.com/morning-joe/watch/boeing-whistleblower...

queueberton Show HN: Comma Separated Values (CSV) to Unicode Separated Values (USV)

No, I have. I don't consider abuse of the format a problem with the format. Though I can see how having to delimit with special characters will help the type of person who writes print(','.join(stuff)).

queueberton Is Cosine-Similarity of Embeddings Really About Similarity?

One does not normally consider trigonometric functions fast to compute. Would you mind elaborating?

queueberton Show HN: Comma Separated Values (CSV) to Unicode Separated Values (USV)

CSV is honestly not that problematic. Figuring out if an field contains and comma and then properly quoting it is trivial. And fields without commas don't need quoting. Sometimes your application even guarantees no commas, especially if CSV is into it from the beginning.

queueberton Roku data breach: Over 15k accounts affected

For years, it has been the case that, when booting up the Roku, the highlighted item would often be a link that would install an app. My kids have accidentally installed so many things. When I tried to remove YouTube, it suggested it below the installed apps, and the kids re-installed it, without having a clue what they were doing and were confused as to why it now showed ads (logged out), when before it didn't (logged in with Premium).

queueberton Webb and Hubble confirm Universe's expansion rate

Somewhere aliens are making fun of how shoddy our simulation is coded.

queueberton The Surprising History of Claims for Life on the Sun (2011)

One generally accepted feature of life is maintaining a localized entropy anomaly, at the expense of dumping excess heat to an outside reservoir. There is no evidence that the Sun is doing anything like that. All of its MHD processes follow normal laws of thermodynamics, where structures will persist at a variety of scales and durations according to universal power-law spectra. The gravitational dynamics of the asteroid belt definitely doesn't exhibit any entropy peaks not explained by Jovian perturbations, etc.

I think you are confusing statistically improbable, but completely normal, patterns with meaningful ones, which is probably the primary flaw of the human brain.

That said, it's also improbable that our meat brains can comprehend every possible form of life in the universe.

queueberton The Surprising History of Claims for Life on the Sun (2011)

Robert Forward wrote a series of sci-fi novels about life on the surface of a neutron star. The first was called Dragon's Egg [1] and is a decent, if tropey, read. Some of the same issues are addressed in his books, of course making a ton of assumptions about the properties of neutron star matter (we still don't know the equation of state for neutron stars).

The idea is very simply that combining abundant energy with something capable of holding structure in such an environment could in theory produce a persistent store of information. A leap, sure, but a fun one.

1. https://en.wikipedia.org/wiki/Dragon%27s_Egg

queueberton FDA clears first over-the-counter continuous glucose monitor

Mathematically, yes. The probabilities involved are different, however.

queueberton FDA clears first over-the-counter continuous glucose monitor

Do you think tearing apart a working airplane and putting it back together in order to inspect it between every flight is a safe practice?

Did you see the 737 door plug accident?

You are missing part of the picture here. And that part is not that everyone else is dumb but you.

queueberton FDA clears first over-the-counter continuous glucose monitor

Doctors make mistakes. Procedures carry inherent risks, such as infection, allergies, blood clots, etc. Even driving to the doctor to discuss your blood glucose carries a risk of car accident. All of these are low probability events, but non-zero.

queueberton FDA clears first over-the-counter continuous glucose monitor

If you knew which signals were false positives, they'd be true negatives.

queueberton FDA clears first over-the-counter continuous glucose monitor

False positives in medicine can cause unnecessary interventions that carry their own risks. Believe it or not, there is an optimum level of ignorance for maximum health.

queueberton Leaving LinkedIn

Since when is LinkedIn is fast paced? It hasn't substantially changed in ten years.

queueberton Why do tree-based models still outperform deep learning on tabular data? (2022)

This is tangential, but that paper has some amazingly good plots for an ML paper.

queueberton Why do tree-based models still outperform deep learning on tabular data? (2022)

Why?

queueberton Why do tree-based models still outperform deep learning on tabular data? (2022)

Not the parent, but NNs typically work better when you can't linearize your data. For classification, that means a space in which hyperplanes separate classes, and for regression a space in which a linear approximation is good.

For example, take the circle dataset here: https://playground.tensorflow.org

That doesn't look immediately linearly separable, but since it is 2D we have the insight that parameterizing by radius would do the trick. Now try doing that in 1000 dimensions. Sometimes you can, sometimes you can't or don't want to bother.

queueberton Java is becoming more like Rust, and I am here for it

Haskell + Ada + '{}' == Rust

queueberton Bookish Diversions: Do Audiobooks Count?

Reading a book exercises a part of the brain that listening does not. And that same part of the brain, the quiet focus, is necessary to solve hard problems and reflect on your choices in life.

queueberton Dune: Part Two Is the Best Sci-Fi Film of the Decade

That's too early -- we'd have no reason to care about him by then. But that would work as the second scene potentially.

queueberton Heinz’s sustainable ketchup cap

Hogsheads are more intuitive.

queueberton Heinz’s sustainable ketchup cap

That used to be part of the excitement.

queueberton Heinz’s sustainable ketchup cap

Are you aware that fruits and vegetables are sold individually without packaging?

queueberton Dune: Part Two Is the Best Sci-Fi Film of the Decade

The Sci-Fi channel Children of Dune miniseries is highly underrated as well.

queueberton Gemini and Google's Culture

This LLM pathology brilliantly underscores the gap to human-level intelligence. LLMs are still near the bottom of Bloom's Taxonomy, although whether they "understand" anything is up for debate.