rcthompson

Hmm, I hadn't considered that doubling the drive size doubles the resilver time and therefore doubles the exposure time for risk of array loss. I guess the math gets complicated depending on RAID topology.

Of note, assuming that decommissioning of drives is driven primarily by e.g. space concerns rather than signs of impending individual drive failures (which seems to be the case based on the linked article about storage scaling), you could conduct a survival analysis in which decommissioned drives are treated as right-censored to get a better measure of the failure rate over time as well as how that failure rate depends on various factors. Note that the most common choice of a proportional hazards model may not be appropriate here, and an accelerated failure time model may be more appropriate, although I couldn't say for sure without actually working with the data.

If hard drives increase in capacity while maintaining the same MTBF, does this count as an improvement? If you previously stored your data on 10 drives and now you can store the same data on 5 drives, that reduces the probability of failure of the system as a whole, right? Is there some kind of "failure rate per byte" measure that normalizes for this?

I thought that it was a log10 scale, so each increment of 1 on the scale is a 10-fold power increase, not a 10^1.5-fold.

In a similar vein, some of my colleagues have been feeding their scientific paper methods sections to LLMs and asking them to implement the method in code, using the LLM's degree of success/failure as a vague indicator of the clarity of the method description.

The author notes that the resulting Rust port is not very "rusty", but I wonder if this could also be solved through further application of the same principle. Something like telling the AI to minimize the use of unsafe etc., while enforcing that the result should compile and produce identical outputs to the original.

DNA methylation means adding one or more methyl groups to the DNA, so technically it is an alteration. But most people would assume that altering a gene specifically means changing the "letters" of the gene sequence that encode the protein, and that's not what DNA methylation does.

I can't speak for every institution, but our PR department does as many back and forth passes as it takes for the scientists who did the work to sign off that any edits made still preserve scientific accuracy.

Describing a change to DNA methylation "alters" a gene is technically correct in the sense that it is an change to the molecular structure of the DNA that makes up the gene, but is indeed misleading, because without further clarification a majority of people would assume it refers to a change in the gene sequence.

Imagine you have the text of a book in a word processor. You can change the text by typing new words or deleting ones that are there. You can also change the font, size, alignment, etc. The latter category of changes does not alter the words in the text, but it can affect how that text is interpreted, which parts of the text a reader focuses on, etc. The difference between a genetic alteration and an epigenetic alteration is conceptually similar. Genetics is changing the "text" of the genome while epigenetics is changing aspects of the genome that affect how that "text" is interpreted and used.

I think they're also confusing power and energy, or at least not clearly distinguishing between them, with phrases like "enough to power X houses".

Are they breaking the law? I don't love that they're requiring either payment or allowing cookies, but at least they're fully disclosing that and giving you the choice. I'm pretty sure that's what the GDPR requires.

Did you roll your own die or just trust that Randall's original die roll was sufficiently random?

There actually are checklists you have to fill out when publishing a paper. You have to certify that you provided all relevant statistics, have not doctored any of your images, have provided all relevant code and data presented in the paper, etc. For every paper I have ever published, every last item on these checklists was enforced rigorously by the journal. Despite this, I routinely see papers from "high-profile" researchers that obviously violate these checklists (e.g.: no data released, a not even a statement explaining why data was withheld), so it seems that they are not universally enforced. (And this includes papers published in the same journals around the same time, so they definitely had to fill out the same checklist as I did.)

Does anyone know where I can see one of these screens in person? I'm interested in potentially owning one for reading comics, graphic novels, etc., but I'd really like to see what it looks like in person before spending money on one.