derefr

63 ms·

571.
by derefr 2y ago
Re: 2 — for the same reason that you shouldn't host your site's status page on the same infrastructure that hosts your site (if people want to see your status page, that probably means your infra is broken), I would guess that Ope
572.
by derefr 2y ago
Kind of feels like it should be a crime for a private party to attempt to write terms into a contract that introduce punishments justified in terms of law, where that justification is based on an interpretation of law that's already be
573.
by derefr 2y ago
It's "early" until the sockets and controllers are standard Shenzhen factory pick-and-place jellybeans. For the first few years of USB-C, it was essentially just being supported by one or two suppliers that most factories did
574.
by derefr 2y ago
For network receive, I was assuming kernel-bypass sockets, not kernel sockets. `bytes` can give you "ring-buffer-like" one -copy kernel-socket receive by e.g. using the Buf as the target for scheduling io_uring read/recv i
575.
by derefr 2y ago
> I'm very well versed in this domain. Apologies, I wasn't really responding to you directly; I was just taking the opportunity to write an educational-blog-post-as-comment aimed at the average HN reader (who has likely never c
576.
by derefr 2y ago
> It's a rather strange crate because network IO doesn't actually need contiguous memory. Network IO doesn't need contiguous memory, no, but each side of the duplex kind of benefits from it in its own way: 1. on receive,
577.
by derefr 2y ago
Yes; in fact, many people "test" this every day, by accident, while trying to set up popular instruction-following models for "roleplaying" purposes, through UIs like SillyTavern. Open models are almost always remotely h
578.
by derefr 2y ago
> memory consolidation happens ~once per year, globally, as past interactions with them almost certainly become future training data. Yes and no. Humans — organic animal brains in general — have a seemingly highly-clever approach to memo
579.
by derefr 2y ago
> The entropy of ChatGPT (as well as all other generative models which have been 'tuned' using RLHF, instruction-tuning, DPO, etc) is so low because it is not predicting "most likely tokens" or doing compression. A LL
580.
by derefr 2y ago
You're conflating two steps: 1. hypothesizing — coming up with a novel insight at runtime, that uncovers new parts of the state space the model doesn't currently reach 2. syllogizing — using an insight you've derived at run
581.
by derefr 2y ago
There were already actual commitments to diversity in most places, yes. DEI programs , on the other hand, were basically a symbolic "party badge" that many companies and organizations felt compelled to adopt to keep scary people
582.
by derefr 2y ago
This doesn't give you the ability to process ideas through the derived new insights, any more than loading the contents of a VLSI program into regular RAM gives you an FPGA. The linear-algebra primitives used in LLM inference, fundam
583.
by derefr 2y ago
I still can't really picture this. (And I've googled the concept to no avail. Is there a high-level overview anywhere? Even one specific to a certain game engine would be nice.) I was initially picturing a DAW VST node graph, wher
584.
by derefr 2y ago
This quote is likely intended for people who've tried other solutions and disliked them, but as someone who's never used a game engine of any kind, I'd appreciate someone giving me an ELI5 of how "nodes" relate to &
585.
by derefr 2y ago
> it's pretty much impossible to bypass as a human if their magical black box doesn't like your browser and/or IP address There are residential-IP-backed VPN services that you can use just like commercial VPN services — bu
586.
by derefr 2y ago
Full anonymity may be valuable, if the set of a paper's reviewers has to stay fixed throughout the review process If peer review worked more like other publication workflows (where documents are handed across multiple teams that revi
587.
by derefr 2y ago
Re: your experiences — racists (and bigots in general) are a lot more likely to voice the bigoted things they're thinking , when the target of their vitriol looks like someone who wouldn't fight back against what they're s
588.
by derefr 2y ago
Yes, racist. I took my (Asian) partner back with me to visit the small town I grew up in, and I could not believe some of the shit coming out of people's mouths when they saw her. Literally 1 in 5 people she talked to while we were the
589.
by derefr 2y ago
When Americans are motivated to move out of the small town they were born in, it's often because that "small town" is a dying and depopulated hollow shell full of awful, racist, jobless, drug-addicted idiots who are mostly on
590.
by derefr 2y ago
If you mean "exactly as architected currently", then yes, current Transformer-based generative models can't possibly be anything other than a dead end. The architecture will need to change at least a little bit, to contin
591.
by derefr 2y ago
The counter-evidence I'll cite comes from language used by criminal investigators in true-crime television like Forensic Files . I recall several instances of language like "we realized that the suspect then disappeared off the g
592.
by derefr 2y ago
> Maybe that challenge question is just a long password (recovery key). Specifically, recovery keys are designed to be so long and complex that no human could possibly memorize them, ensuring that they’re written down / printed inst
593.
by derefr 2y ago
ChatGPT knows about other domains (e.g. software versions) where that inequality is true. Try telling it you’re doing arithmetic.
594.
by derefr 2y ago
I would agree that at least some OF consumers that throw donations at the OF creators, are knowingly buying an illusion. But I don't think your analogy holds. I think there's a fundamental difference between these two activities (
595.
by derefr 2y ago
> Yes, in an ideal world, none of this would be a problem. But that's not the world we have right now, and I don't believe that minor modifications of this concept of device can improve the state of it. I've been trying fo
596.
by derefr 2y ago
> what is the actual good-faith rationale for using this feature So, responding to viewers increases engagement, and thereby a channel's virality. And one human can only do so much of that. So eventually, you hit a marketing scalab
597.
by derefr 2y ago
Presumably the goal of the people throwing money at an OF content-creator [in the form of "donations" — "here's some extra money without any implied obligation" — rather than e.g. paying for custom content] is to tr
598.
by derefr 2y ago
Why'd you switch away from Cloudflare['s registrar]?
599.
by derefr 2y ago
> The BrandShield software is probably instructed to eradicate all "unauthorized" use of their trademark, so they sent reports independently to our host and registrar claiming there was "fraud and phishing" going on,
600.
by derefr 2y ago
Very likely correct in the literal sense (you shouldn’t rely on the published software); but I believe the approach it uses is still relevant / generalizable. I.e. you can take whatever the current state-of-the-art facial recognition

More ›