derefr

106 ms·

841.
by derefr 2y ago
I imagine you could "solve" this (for a painful and pointless value of "solve") by 1. only allowing the Service Worker to do constant-time versions of operations (like the constant-time primitives that cryptographic code
842.
by derefr 2y ago
If a page can already deduce performance fluctuations all on its own, then you don't need a special access-limited performance API, do you? Just have the page do whatever you're imagining could be done to extract this side-chann
843.
by derefr 2y ago
> How would you make an API accessible on the browser side but prevent the return values from being sent to the server? Create an API for starting a "performance-metrics visualization Service Worker", that takes two things from
844.
by derefr 2y ago
I mean, you can and you can't. AFAIK, the UEFI spec imposes no requirement that (non-hotplug) devices be re-initializable after you've already initialized them once. Devices are free to take the "ExitBootServices has been cal
845.
by derefr 2y ago
You're right (as I saw another comment cite the primary-source for); but I'm still curious now, whether there'd be a way to pull this off. > I think that would mean you cannot use native drivers Yes, that's right. &gt
846.
by derefr 2y ago
> instead of immediately loading the System A userland I mean, it kind of is loading the System A userland. At least the initramfs of it. AFAICT in the proposal the bootloader would now be a regular userland program living in the initr
847.
by derefr 2y ago
The sibling comments who think you need to jump back to EFI to solve this, are thinking in layer-ossified terms. This is Redhat proposing this, and they're perfectly confident in upstreaming kernel patches to make this happen. I would
848.
by derefr 2y ago
I was picturing doing the deduplication as a map-reduce process over a huge (petabytes) dataset, where every worker is blind to what embeddings other workers have already generated. In such a case, shoving the embeddings generated by each w
849.
by derefr 2y ago
If this is true, then why do you think that — as the OP article states — the developers of GPT3 chose to use non-ML-based techniques to deduplicate their dataset, when they would be the most equipped to use ML-based approaches? Just the p
850.
by derefr 2y ago
You might need to DIY, but "mad scientist" is doing a lot of work in your statement that I don't think applies. You just need to follow some very-thoroughly-detailed online tutorials. It's one of those things like baking
851.
by derefr 2y ago
From https://en.wikipedia.org/wiki/Resource_fork : > While the data fork allows random access to any offset within it, access to the resource fork works like extracting structured records from a database. So, whateve
852.
by derefr 2y ago
As a document clustering / dataset deduplication technique, how well does "throwing ML at the problem" (e.g. using a pre-trained LLM encoder to generate vector embeddings of your documents, and then sticking those vectors int
853.
by derefr 2y ago
> Just check out Common Crawl if you want to get an idea of what it would look like. It has a lot of rubbish, sure, but the reason that that matters with Common Crawl, is that Common Crawl isn't a continuous stream; it's rath
854.
by derefr 2y ago
> which is kind of annoying since he's the most active C#/.Net recruiter in my area Tangent: I like .NET as a platform, but I get the impression that a lot of .NET shops tend to be toxic in this particular way. .NET attracts bi
855.
by derefr 2y ago
Given the 15 year gap between the events and this post about them, I'm pretty sure OP isn't still angry. They likely were just working on a search engine, and this old story came to mind, and they realized that it'd be perfec
856.
by derefr 2y ago
> I'm wondering the same thing. Well, if you really want to know, you could try taking the HTTP responses for the page you expect to be highly ranked, and the page that's actually highly-ranked, and applying various common rank
857.
by derefr 2y ago
But regularly solving problems outside of your expertise is a necessary job duty in pretty much any software job. (If I were conducting the interview, I might explicitly explain that I'm going to ask them a few questions with the goa
858.
by derefr 2y ago
> It’s you who chooses what sites we crawl Yeah, but you still reserve the right to not crawl sites (or to remove them from your index), yes? So there's still the opportunity to do evil. I'm still waiting for a "raw&qu
859.
by derefr 2y ago
> How many young children do you genuinely think would do problems like that without messing up a step before having drilled for quite some time? You're overestimating how many examples "drilled for quite some time" repres
860.
by derefr 2y ago
I assume you know this, but depending on what the enclosure was made of, what frequencies the devices inside were broadcasting at, and where you placed them inside, you could very well have accidentally produced a resonating chamber, parabo
861.
by derefr 2y ago
> This is quite plainly nonsense. Every other manufacturer allows WiFi adapter replacements because every module on the market must be individually FCC certified. Perhaps every other OEM are running antennas through the case that have ex
862.
by derefr 2y ago
Why are young children able to quickly surpass state-of-the-art ML models at arithmetic tasks, from only a few hours of lecturing and a "training dataset" (worksheets) consisting of maybe a thousand total examples? What is happeni
863.
by derefr 2y ago
Asking questions about a document is one thing; asking questions that synthesize information across many documents — the human-intelligent equivalent of doing a big OLAP query with graph-search and fulltext-search parts on your email databa
864.
by derefr 2y ago
> There is a real danger that an LLM will include information in the summary that does not come from the original email thread, but from other emails the model was trained on. You could learn from the summary that you are going to get fi
865.
by derefr 2y ago
No, it makes sense: he's coming at it from the perspective of knowing exactly what task you want to accomplish (something like "fixing the grammar in this document.") In such cases, a model only has to be sufficiently smart
866.
by derefr 2y ago
Tangent: I would note that a "password protect an arbitrary link" service could actually be made into a "thin, just-in-time" form of one of these services... by not being a redirect service, but rather being a dynamic,
867.
by derefr 2y ago
You don't hear what they're saying. On HN, and especially regarding "infrastructure" services that you might depend on indefinitely for something, "what's the business model" isn't a question about ho
868.
by derefr 2y ago
I would point out that this will likely be used 99% of the time by people who want to charge for access to content... that they don't themselves own. If you've ever wandered the skeevier parts of the internet (the parts that a tee
869.
by derefr 2y ago
I wasn't so much imagining each shell script being replaced with a binary, as I was imagining deploying a single static binary "sysadmin-DSL interpreter binary" — where that "interpreter" is just "Ruby with a c
870.
by derefr 2y ago
Yes, in theory, any filesystem could trivially add a feature of "ref-counted immutable extents" — where a special syscall equivalent to `cat a b c > d` could be implemented that creates an inode d that consists of references to

More ›