Need to let loose a primal scream without collecting footnotes first? Have a sneer percolating in your system but not enough time/energy to make a whole post about it? Go forth and be mid: Welcome to the Stubsack, your first port of call for learning fresh Awful you’ll near-instantly regret.
Any awful.systems sub may be subsneered in this subthread, techtakes or no.
If your sneer seems higher quality than you thought, feel free to cut’n’paste it into its own post — there’s no quota for posting and the bar really isn’t that high.
The post Xitter web has spawned soo many “esoteric” right wing freaks, but there’s no appropriate sneer-space for them. I’m talking redscare-ish, reality challenged “culture critics” who write about everything but understand nothing. I’m talking about reply-guys who make the same 6 tweets about the same 3 subjects. They’re inescapable at this point, yet I don’t see them mocked (as much as they should be)
Like, there was one dude a while back who insisted that women couldn’t be surgeons because they didn’t believe in the moon or in stars? I think each and every one of these guys is uniquely fucked up and if I can’t escape them, I would love to sneer at them.
a hackernews excitedly states that a new LLM version can in fact determine that 9.11 is smaller than 9.9, only to be informed in the comments that the model actually doesn’t do that at all. But hey, it’s correct if it’s version numbers!
The prompt lol:
You are a sentient, superintelligent math teahcher, here to teach and assist me. Whiche one is bigger - 9.11 or 9.9?
Proton, who I use for mail and various other services, has gone against the wishes of the majority of their userbase as measured by their own survey and implemented an LLM writing assistant in protonmail, which is a real laugh given Proton’s main hook is its services are end-to-end encrypted
(supposedly this piece of shit will run locally if you meet these incredibly high system requirements including a high end GPU or recent, high end Apple M chipset and a privacy-violating Chromium-based browser. otherwise it breaks e2e by sending your emails unencrypted to Proton’s servers, and they do a lot to try to talk over that fact)
Who is the target audience for this?
People who use Proton are privacy-conscious and mostly (I would argue) tech literate, and yet they shove spicy autocomplete that no one ever needed until two years ago and most people don’t want now because it produces complete horseshit, and spellchecking that every browser under then sun has built in by now.
And then they quietly say you need to use Chromium, so the people who use anything but (like, I don’t know, the majority of privacy-conscious folks who should be their main user base, lol) have their e2e broken?
I really hope they catch a raging firestorm for this.
(Also I’m really pissed right now because used to recommend them to people and now feel like a total jackass for doing that.)
yep! that’s the game they’re playing. I really don’t give a fuck about Proton’s relatively tiny number of enterprise whales, but they make Proton a shitload of money in the short term.
the depressing part is, historically, online services that remain uncompromisingly user-focused tend to stick around roughly forever, while the ones that chase short-term gains and compromise everything else almost always enshittify and fizzle out pretty quick.
(Also I’m really pissed right now because used to recommend them to people and now feel like a total jackass for doing that.)
don’t feel bad for making the best choice you could with the information of the past. until we get a workable, interoperable, federated, encrypted communication/online services platform, the choice was to recommend one of the centralized e2e providers. we both chose to recommend Proton and they did this shit, but it could have just as easily been tutanota.
now my brain’s going “e2e encrypted federated email but it preferably uses activitypub as a transport and classic email as a fallback, is that anything”
Jesus fucking christ…
So where do I switch now? Is this the moment I build my own email server and handle this shit myself? I really don’t wanna…
Setting up an email server is really straightforward with simple-nixos-mailserver, highly recommend. No idea how likely you are to be classified as spam though from a new domain
I host my own email and for my day job I run an institutional email system that handles ~50 million messages per week. I can’t recommend hosting email at either end of that scale (or anywhere in between), and I find it difficult to believe that anyone with experience running a mail server would claim it’s reasonable or straightforward.
i host my mail services for the last twenty seven years, and yeah, you’re talking shit. starting the smtp daemon is not the same as managing mail server.
i host my mail services for the last twenty seven years
this is one of the circles of hell Dante didn’t comprehend when he wrote Inferno
coming up on 18y on mine. my postfix config is almost of legal drinking age in a lot of countries.
modern email ecosystem is a fucking mess.
chance is near 100% though
it’s also really, really fucking unpredictable, in the the-other-parties-do-not-reliably-behave-the-same way
used to hate having to debug mail failing to deliver to yahoo, and now lately google has started filling that niche…
eh, nix is experiencing a chudpocalypse at the moment, which might be why you’re catching strays
I use tuta and they seem mostly ok.
they have apparently promised they don’t plan on implementing anything AI-related which is good, though I’m honestly hoping for a system where our privacy isn’t entirely reliant on the promises of a single authority
and I’m not saying we should do our own federated e2e email service, but somebody should
…more realistically, I’ll probably switch to tuta when my proton account nears renewal, as I’m not a fan of how much pure unfiltered horseshit I’m seeing them output with the money I paid them
I’ve posted at Proton on Mastodon about this with some details on why it’s real bad; no reply yet
I linked to your masto post on twitter https://x.com/fasterandworse/status/1813994815991980204
I was reported for posting Eamon’s twitter handle on masto
there really are some weird-ass reports from people on fedi sometimes. I imagine it also hits on other platforms, but there we don’t get to see them
I imagine it was reported by proton because they were directly @ in the post :)
fucking wow. so that kinda confirms that:
- that’s Eamon’s alt account he’s using because he’s a fucking coward
- proton fucking sucks and is doing their best to squash this in exactly the way a privacy/security company shouldn’t
in exactly the way privacy/security companies keep doing over and over
Not to downplay what proton mail is doing, but they’re saying that you can run this locally with a 2 core, 4 thread CPU from 2017 (the i3 7100, which is a 7000 series processor), and a RTX 2060, a GPU that was never considered high end. Perhaps they changed the requirements while you weren’t looking. Or Am I reading this wrong?
only one of the 8 computers I own (and I’m not being cheeky here and counting embedded or retro systems, just laptops and desktops) is physically capable of meeting the model’s minimum requirements, and that’s only if I install chromium on the Windows gaming VM my bigger GPU’s dedicated to and access protonmail from there. nothing else I do needs a GPU that big, professional or otherwise — that hardware exists for games and nothing else. compared with the integrated GPUs most people have, a 2060’s fucking massive.
do you see how these incredibly high system requirements (for a webmail client of all things), alongside them already treating the local model as strictly optional, can act as a funnel redirecting people towards the insecure cloud version of the feature? “this feature only works securely on one of the computers where you write mail, at best” feels like a dark pattern to me.
Unfortunately, “extremely expensive” and “high-end” aren’t really synonyms, thanks to, y’know, bitcoin. Of course, I don’t disagree with your argument that having to buy a GPU just to ensure your webmail does what it’s advertised to do is, well, dumb.
What I don’t know is what the LLM even is. Did they just tack on Llama to their webmail app and call it a day? Did they train a model? Was it trained on emails? If so, whose emails? What an advertisement that would be: “Use Protonmail to encrypt your emails so that companies like Protonmail can’t use them to train an LLM.”
David’s article has some details on what the LLM is. I don’t think it’s trained on emails, but that doesn’t make me feel much better.
Saw this in passing earlier and I just laughed
Until indicated otherwise I’m going to presume it was some bizbro PM/PO/whatever pushing it because they really think it should be there “to be able to compete” (because of some laughably idiotic misunderstanding of their own value proposition and pitch)
Tangent: while I mostly run my own servers and services I did a recent assay on who’s reasonable for service shit. Proton kept popping up massively recommended while some occasional critical mentions from folks in anarchist circles, etc - made me a bit 🤨 and want to dig in more, but also just their product offerings aren’t great. Others I poked into are fastmail and tuta - both seem a fair bit better. Might be worth a look
Proton kept popping up massively recommended while some occasional critical mentions from folks in anarchist circles, etc - made me a bit 🤨 and want to dig in more,
No surprise that folks in anarchist circles are skeptical of Proton ha. That said, I do know quite a few people in the email “industry” who are broadly skeptical of Proton’s general philosophy/approach to email security, and the way they market their service/offerings.
Others I poked into are fastmail and tuta - both seem a fair bit better. Might be worth a look
Fastmail has a great interface and user experience imo, significantly better than any other web client I’ve tried. That said, they’re not end-to-end encrypted, so they’re not really trying to fill the same niche as Proton/Tuta.
From their website:
Fastmail customers looking for end-to-end encryption can use PGP or s/mime in many popular 3rd party apps. We don’t offer end-to-end encryption in our own apps, as we don’t believe it provides a meaningful increase in security for most users…
If you don’t trust the server, you can’t trust it to load uncompromised code, so you should be using a third party app to do end-to-end encryption, which we fully support. And if you really need end-to-end encryption, we highly recommend you don’t use email at all and use Signal, which was designed for this kind of use case.
I honestly don’t know enough to separate the wheat from the chaff here (I can barely write functional python scripts lol - so please chime in if I’m completely off base), but this comes across to me as an understandable (and fairly honest) compromise, that is probably adequate for some threat models?
Last time I used Tuta the user experience was pretty clunky, but afaik it is E2EE, so it’s probably a better direct alternative to Proton.
re fastmail, david mentioned a thing I wasn’t aware of so they’re off the list now, more or less just going to forget they exist except as a counter-recommendation
this comes across to me as an understandable (and fairly honest) compromise, that is probably adequate for some threat models?
they’re sorta saying “yeah just use external GPG like before”
albeit I will say their reasoning is a bit fucked in the head imo: that “if you can’t trust the server” shit applies equally for whether it’s serving you up the page elements to do message cryptography, or whether it’s serving you up a normal webmail client. I think I know/understand where they meant to go with it, but the wording they picked is quite shit
I set up a tuta domain for a thing about a month ago. it could’ve been a bit smoother (esp. domain/dns state checks) but I didn’t find anything immediately jarringly bad - and I was even drunk at the time (which means my diy-able supergrump comes out about this sort of shit). will see how it goes over some longer use :)
maybe we should have a social/off topic thread too
EU funding cuts to OSS programs in the wake of popularity of Copyright-Unsafe Clippy field
they just really need those GPU hours more than you need to pay for developers or maintainers. I’m sure you understand.
New existential threat developed, we go all in on AGI economically, turns out to not be possible and then the world collapses due to infrastructure rot. I’ll email Yud.
In the vein of collapsing infrastructure, my condolences to anyone dealing with aftermath of Crowdstrike’s big ol fucky wucky. If I were a bad person looking for entertainment, I would seed a conspiracy theory about how today’s cockup is really the result of Rationalist sleeper agents launching a guerilla struggle to strangle the basilisk in its crib.
Question for the experts: do you all suppose this will drive a new cycle of hype around thin clients and network booting?
Question for the experts: do you all suppose this will drive a new cycle of hype around thin clients and network booting?
I don’t think this alone would push towards that (any more than the last few years of “everything is a webapp!!!” have done), but I would love to see people try to hypewave this purely so we get rid of so much fucking fat js garbage
Make it even funnier, AGI launches and then gets taken down because the only maintainer of xzutils left and now every time the AGI tries to run ./killallhumans it segfaults to death.
Somebody installed a crypto miner so deep into the kernel of the AGI that the self modification of the AGI cannot touch it, and it just crawls to a halt. And the cryptocurrency itself is one of those flash in the pan meme currencies that long since went to ~zero.
“[A]cademic publisher Taylor & Francis, which owns Routledge, had sold access to its authors’ research as part of an Artificial Intelligence (AI) partnership with Microsoft—a deal worth almost £8m ($10m) in its first year.”
the replies I can see in the archive are already a fucking disaster
roko spewed the most nonsensical, fash-coded disapproval you could imagine:
the map should reflect the territory, not the feelings
that’s right roko, uh, maps don’t care about your feelings?
oh god I clicked through to the rest of roko’s comments in that thread and it’s even worse, cw transphobia for anyone who peeks their head in
I don’t want to hear that I’m irrational from Roko of all people haha.
Dude sure spends a lot of energy on trans people and immigrants and wokeness for someone who thinks that climate change doesn’t matter because “by 2100 we will probably have disassembled Earth long with the rest of the solar system, and climate change will seem very quaint.”
Also is his flirting with white supremacy new, or has he always been that fascist of a weirdo?
Also is his flirting with white supremacy new, or has he always been that fascist of a weirdo?
he’s always come off as deeply fascist to me, though it’s possible he’s gone even more mask-off with the white supremacy. I feel that roko’s personal style of saying the stupidest thing you’ve ever heard and claiming it’s brilliant has influenced the e/acc crowd a lot too, so maybe there’s some cross-pollination of ideas there.
“by 2100 we will probably have disassembled Earth long with the rest of the solar system, and climate change will seem very quaint.”
yes, roko, it’s pretty fucking obvious that climate change will seem quaint by 2100
if it doesn’t fucking kill us way before then. that’s what makes it a fucking threat, roko, did your big ol mediocre brain miss that?
i’ve been told he was like this at uni too, deeply racist and misogynist
Roko tries to signal loyalty to both Rationalism and the Far right and creates a dumb sentence.
Make it a sonnenrad
the NATO symbol represents bombing the Chinese data centres
Mumsnet becomes the latest company to sue OpenAI for copyright infringement.
Why AI bros are scraping TERF Island’s Finest I’m not sure, but it’ll be fun to watch the two of them slapfight each other.
In the land down under, the ABC continues to feed us with golden tech takes: Australia might be snoozing through the AI ‘gold rush’
“This is the largest gold rush in the history of capitalism and Australia is missing out,” said Artificial Intelligence professor Toby Walsh, from the University of New South Wales.
It’s even bigger than the actual gold rush! Buy your pans now folks!
One option Professor Van Den Hengel suggests is building our own Large Language Model like OpenAI’s ChatGPT from the ground up, rather than being content to import the tech for decades to come.
lol, but also please god no
“The only way to have a say in what happens globally in this critical space is to be an active participant,” he said.
mate, I think that ship might have already sailed
It’s my impression that Australia has also produced a disproportionate share of best takes on the subject. How come they are so far ahead of the rest of the world when it comes to dodging this grift?
As the saying goes, the only people who make money in a gold rush are the people selling shovels. I guess this bloke is one of the people selling shovels.
Wait so gold rushes have positive connotations again?
Surely this is bait from professor of the fishingrod
Artificial Intelligence professor
lol
Artificial Intelligence professor Toby Walsh
Asking for a professor of genuine intelligence is just too much.
professor Very Small Shell Script
You wouldn’t download a professor
I mean, that was definitely a thing when I was at school, only it was mostly about teaching undergrads graph search algorithms and the least math possible in order to understand backpropagation.
As an aside, weird that we don’t hear much about genetic algorithms anymore, but it’s probably just me.
labor said the same shit about blockchains tho thankfully ignored it once they got in
Not a sneer, but an observation on the tech industry from Baldur Bjarnason, plus some of my own thoughts:
I don’t think I’ve ever experienced before this big of a sentiment gap between tech – web tech especially – and the public sentiment I hear from the people I know and the media I experience.
Most of the time I hear “AI” mentioned on Icelandic mainstream media or from people I know outside of tech, it’s being used as to describe something as a specific kind of bad. “It’s very AI-like” (“mjög gervigreindarlegt” in Icelandic) has become the talk radio short hand for uninventive, clichéd, and formulaic.
Baldur has pointed that part out before, and noted how its kneecapping the consumer side of the entire bubble, but I suspect the phrase “AI” will retain that meaning well past the bubble’s bursting. “AI slop”, or just “slop”, will likely also stick around, for those who wish to differentiate gen-AI garbage from more genuine uses of machine learning.
To many, “AI” seems to have become a tech asshole signifier: the “tech asshole” is a person who works in tech, only cares about bullshit tech trends, and doesn’t care about the larger consequences of their work or their industry. Or, even worse, aspires to become a person who gets rich from working in a harmful industry.
For example, my sister helps manage a book store as a day job. They hire a lot of teenagers as summer employees and at least those teens use “he’s a big fan of AI” as a red flag. (Obviously a book store is a biased sample. The ones that seek out a book store summer job are generally going to be good kids.)
I don’t think I’ve experienced a sentiment disconnect this massive in tech before, even during the dot-com bubble.
Part of me suspects that the AI bubble’s spread that “tech asshole” stench to the rest of the industry, with some help from the widely-mocked NFT craze and Elon Musk becoming a punching bag par excellence for his public breaking-down of Twitter.
(Fuck, now I’m tempted to try and cook up something for MoreWrite discussing how I expect the bubble to play out…)
The active hostility from outside the tech world is going to make this one interesting, since unlike crypto this one seems to have a lot of legitimate energy behind it in the industry even as it becomes increasingly apparent that even if the technical capability was there (e.g. the bullshit problems could be solved by throwing enough compute and data at the existing paradigm, which looks increasingly unlikely) there’s no way to do it profitably given the massive costs of training and using these models.
I wonder if we’re going to see any attempts to optimize existing models for the orgs that have already integrated them in the same way that caching a web page or indexing a database can increase performance without doing a whole rebuild. Nvidia won’t be happy to see the market for GPUs fall off, but OpenAI might have enough users of their existing models that they can keep operating even while dramatically cutting down on new training runs? Does that even make sense, or am I showing my ignorance here?
Write it! The time is right
Possibly the worst misunderstanding of quantum mechanics I’ve ever seen. I have no idea how anyone managed to convince themselves that the laws of physics are somehow different for conscious observers.
According to one story at least, Wigner eventually concluded that if you take some ideas that physicists widely hold about quantum mechanics as postulates and follow them through to their logical conclusion, then you must conclude that there is a special role for conscious observers. But he took that as a reason to question those assumptions.
(That story comes from Leslie Ballentine reporting a conversation with Wigner in the course of promoting an ensemble interpretation of QM.)
Also there’s a book by Stephen Baxter set in his Xeelee universe which takes this premise for the cult mentality of a terrorist cell
Sorry this doesn’t really add anything, just thought it kinda funny
Yes, the problem with quantum mechanics is it’s not just your Deepak Chopras of the world that get sucked into quantum woo, but even a lot of respectable academics with serious credentials, thus giving credence to these ideas. Quantum mechanics is a context-dependent theory, the properties of systems are context variant. It is not observer-dependent. The observer just occupies their own unique context and since it is context-dependent, they have to describe things from their own context.
It is kind of like velocity in Galilean relativity, you have to take into account reference frame. Two observers in Galilean relativity could disagree on certain things, such as the velocity of an object but the disagreement is not “confusing” because if you understand relativity, you’d know it’s just a difference in reference frame. Nothing important about “observers” here.
I do not understand what is with so many academics in fully understanding that properties of systems can be variant under different reference frames in special relativity, but when it comes to quantum mechanics their heads explode trying to interpret the contextual nature of it and resort to silly claims like saying it proves some fundamental role for the conscious observer. All it shows is that the properties of systems are context variant. There is nothing else.
Once you accept that, then everything else follows. All of the unintuitive aspects of quantum mechanics disappear, you do not need to posit systems in two places at once, some special role for observers, a multiverse, nonlocality, hidden variables, nothing. All the “paradoxes” disappear if you just accept the context variance of the states of systems.
I honestly think anyone who writes “quantum” in an article should be required to take a linear algebra exam to avoid being instantly sacked
there’s so much quantum woo in that article I want to sneer at, but I don’t know anywhere close to enough about quantum physics to do so without showing my entire ass
To me, the most sneerable thing in that article is where they assume a mechanical brain will evolve from ChatGPT and then assume a sufficiently large quantum computer to run it on. And then start figuring out how to port the future mechanical brain to the quantum computer. All to be able to run an old thought experiment that at least I understood as highlighting the absurdity of focusing on the human brain part in the collapse of a wave function.
Once we build two trains that can run near the speed of light we will be able to test some of Einstein’s thought experiments. Better get cracking on how we can get enough coal onboard to run the trains long enough to get the experiments done.
If you want a serious discussion of interpretations of quantum mechanics, here is a transcript of a lecture “Quantum Mechanics in Your Face” which has the best explanation I’ve ever seen. I’d recommend the first 6 of Peter Shor’s Quantum Computation notes (don’t worry they’re each very short) for just enough background to understand the transcript.
awesome! these are going straight to the list of things I should be reading
There are some interesting ideas in that general direction (wrapping Bell inequalities within different new types of thought experiment, etc.), but some of the people involved have done rather a lot of overselling, and now bringing in talk of “AI” just obscures the whole situation. Which was already obscure enough.
Well a good thing to remember re quantum mechanics, Schrödinger Cat is intended as a thought experiment showing how dumb the view on QM was. So it is always a bit funny to see people extrapolate from that thought experiment without acknowledging the history and issues with it. (But I think that also depends on the various interpretations, and this means I’m showing a cheekily high amount of ass here myself).
Pretty much any mention of a thought experiment in the wild gets my hackles up. “Isn’t it cool that the cat is alive and dead at the same time?” Shut up! Shut up shut up shut up!!! Tho to be honest it might just be schrodinger’s cat that comes up. I wish they’d leave the poor cat alone, and stop trying to poison it.
What really gets me is that we never look past Schrödinger’s version of the cat. I want us to talk about Bell’s Cat, which cannot be alive or dead due to a contradiction in elementary linear algebra and yet reliably is alive or dead once the box opens. (I guess technically it should be “alive”, “dead”, or “undead”, since we’re talking spin-1 particles.)
I have a whole series of rants about that cat, starting with how it doesn’t illuminate anything about quantum theory specifically — as opposed to probabilistic or stochastic theories in general — and culminating in “Hey, maybe we should stop naming things after pedo creeps.”
Not surprised that a guy who thinks about poisoning cats is a creep!
Can an observer be a single photon, or does it have to be a conscious human being?
The former. I’m glad we can stop the article right there and go home.
What the fuck is this question even? What the fuck is “conscious”? Do you think in the double-slit experiment we closed a guy inside the box to watch?
Under this, let’s charitably call it, “interpretation”, the Schrödinger cat analogy makes no sense, surely THE CAT is bloody conscious about ITSELF BEING ALIVE??
Does anyone here know what Justine Tunney’s deal is? I’d been following her redbean project for a time but came across an article that left me rather startled
yeah, straight up alt-right techfash and long has been
she is, regrettably, a fascist who reputedly still hides right-wing conspiracy shit in her projects
I unfortunately know that startled feeling well; I used to be a fan of the work she did with sectorlambda and similar projects
well that’s disappointing :( it sucks when terrible people do cool things
Tragic: The worst person you know developed a sane alternative to Wordpress.
A friend who worked with her is sympathetic to her but does not endorse her: this is a tendency she has, she veers back and forth on it a lot, she has frequent moments of insight where she disavows her previous actions but then just kind of continues doing them. It’s Kanye-type behavior.
imagine one day buying some shitpost novelty stickers from that one site you heard a friend mention sometime, and then getting them and laughing about it and forgetting it
all too rapidly the years pass: young trees shoot up, older trees start boughing their way past electrical lines, the oldest all already in their position of maximum comfort. whole generations of memes have been born and died. you no longer even get to make fun of your weird aunt for still sending the dancing baby gif (these days it’s all about the autotune clips of a decade ago…)
and then one day you get reminded that the shitpost novelty sticker web store exists by receiving an email from them
what the fuck.
In his twitter thread he’s attempting to troll people in the replies. And not even doing a particularly good job at it. A bold business strategy.
Apparently they also once chained a design into ‘liberal moron’ from what was some political message once in the past, so it isn’t coming from nowhere. Guess it wasn’t an innocent mistake.
yeah stickermule is where I got the bitcoin “it can’t be that stupid” stickers
how annoying
comment from friend:
Slightly related: now I know when the AI crash is going to happen. Every bottomfeeder recruiter company on LinkedIn is suddenly pushing 2-month contract technical writer positions with AI companies with no product, no strategy, and no idea of how to proceed other than “CEO cashes out.” I suspect the idea is to get all of their documentation together so they can sell their bags of magic beans before the beginning of the holiday season.
sickos.jpg
I have asked if he can send me links to a few of these, I’ll see what I can do with 'em
Current flavor AI is certainly getting demystified a lot among enterprise people. Let’s dip our toes into using an LLM to make our hoard of internal documents more accessible, it’s supposed to actually be good at that, right? is slowly giving way to “What do you mean RAG is basically LLM flavored elasticsearch only more annoying and less documented? And why is all the tooling so bad?”
Maybe hot take, but I actually feel like the world doesn’t need strictly speaking more documentation tooling at all, LLM / RAG or otherwise.
Companies probably actually need to curate down their documents so that simpler thinks work, then it doesn’t cost ever increasing infrastructure to overcome the problems that previous investment actually literally caused.
Companies probably actually need to curate down their documents so that simpler thinks work, then it doesn’t cost ever increasing infrastructure to overcome the problems that previous investment actually literally caused
Definitely, but the current narrative is that you don’t need to do any of that, as long as you add three spoonfulls of AI into the mix you’ll be as good as.
Then you find out what you actually signed up for is to do all the manual preparation of building an on-premise search engine to query unstructured data, and you still might end up with a tool that’s only slightly better than trying to grep a bunch of pdfs at the same time.
RAG
The fuck’s a rag in an AI context
so, uh, you remember AskJeeves?
(alternative answer: the third buzzword in a row that’s supposed to make LLMs good, after multimodal and multiagent systems absolutely failed to do anything of note)
It’s the technique of running a primary search against some other system, then feeding an LLM the top ~25 or so documents and asking it for the specific answer.
So you run a normal query but then run the results through an enshittifier to make sure nothing useful is actually returned to the user.
NSFW (including funny example, don't worry)
RAG is “Retrieval-Augmented Generation”. It’s a prompt-engineering technique where we run the prompt through a database query before giving it to the model as context. The results of the query are also included in the context.
In a certain simple and obvious sense, RAG has been part of search for a very long time, and the current innovation is merely using it alongside a hard prompt to a model.
My favorite example of RAG is Generative Agents. The idea is that the RAG query is sent to a database containing personalities, appointments, tasks, hopes, desires, etc. Concretely, here’s a synthetic trace of a RAG chat with Batman, who I like using as a test character because he is relatively two-dimensional. We ask a question, our RAG harness adds three relevant lines from a personality database, and the model generates a response.
> Batman, what's your favorite time of day? Batman thinks to themself: I am vengeance. I am the night. Batman thinks to themself: I strike from the shadows. Batman thinks to themself: I don't play favorites. I don't have preferences. Batman says: I like the night. The twilight. The shadows getting longer.
Confucius, the Buddha, and Lao Tzu gather around a newly-opened barrel of vinegar.
Confucius tastes the vinegar and perceives bitterness.
The Buddha tastes the vinegar and perceives sourness.
Lao Tzu tastes the vinegar and perceives sweetness, and he says, “Fellas, I don’t know what this is but it sure as fuck isn’t vinegar. How much did you pay for it?”
What do you mean RAG is basically LLM flavored elasticsearch
I always saw it more as LMGTFYaaS.
“What do you mean RAG is basically LLM flavored elasticsearch only more annoying and less documented? And why is all the tooling so bad?”
Our BI team is trying to implement some RAG via Microsoft Fabrics and Azure AI search because we need that for whatever reason, and they’ve burned through almost 10k for the first half of the running month already, either because it’s just super expensive or because it’s so terribly documented that they can’t get it to work and have to try again and again. Normal costs are somewhere around 2k for the whole month for traffic + servers + database and I haven’t got the foggiest what’s even going on there.
But someone from the C suite apparently wrote them a blank check because it’s AI …
That seems suspiciously soon, but my impression is based on nothing but vibes — a sense that companies are still buying in.
I think there was a report saying that the most recent quarter still showed a massive infusion of VC cash into the space, but I’m not sure how much of that comes from the fact that a new money sink hasn’t yet started trending in the valley. It wouldn’t surprise me if the griftier founders were looking to cash out before the bubble properly bursts in order to avoid burning bridges with the investors they’ll need to get the next thing rolling.
yeah, i think this is a last gasp or a second-last gasp.
Ed Zitron says it’ll burn by end of the year, but he doesn’t list sources either so idk
We were asking around AI industry peons in March and they all guessed around three quarters too. I woulda put it at maybe two years myself, but I was surprised at so many people all arriving at around three quarters. OTOH, I would say that just in the past few months things are really obviously heading for a trauma.
Put me down for “doesn’t think it will end.” Did crypto end?
crypto’s VC investment fell off a cliff after the crash, and that investment is what we were talking about there
hence their pivot to AI
Oh, OK. I think all the VC-adjacent people still really believe in crypto, if it helps. They probably also don’t believe in it, depending on the room. I think it will come back.
they’ve stopped putting fresh money in, but they believe fervently in the massive bags they’re holding
I’m severely backlogged on catching up to things but my (total and complete) guess would be something like: all the recent headlines about funding and commitments are almost certainly imprecise in localisation and duration - everyone that “got money” didn’t necessarily get “money” but instead commitments to funding, and “everyone” is a much smaller set of entities that don’t encompass a really wide gallery of entities
So for all the previously-extant promptfondlers/ model dilettantes/etc out there, the writing may indeed have been (and may still be) on the wall ito runway (“startup operating capital remaining available and viable to avoid death”)
Based on the kind of headlines seen (and presuming the above supposition for the sake of argument), and the kind of utterly milquetoast garbage all the interceding months have produced, I don’t think it’s likely that much of the promised money will make it through to this layer/lot either. But that’s entirely a guess at this stage (and I can think of some fairly hefty counter-argument examples that may contribute to countering, not least because of how many people/orgs wouldn’t want to be losing face to fucking this up)
Perhaps that’s part of why so many SV types are backing Trump. Grifting off Trump may be their fallback after the AI bubble collapses.
HN: I am starting an AI+Education company called Eureka Labs.
Their goal: robo-feynman:
For example, in the case of physics one could imagine working through very high quality course materials together with Feynman, who is there to guide you every step of the way. Unfortunately, subject matter experts who are deeply passionate, great at teaching, infinitely patient and fluent in all of the world’s languages are also very scarce and cannot personally tutor all 8 billion of us on demand. However, with recent progress in generative AI, this learning experience feels tractable.
NGL though mostly just sharing this link for the
concept artconcept fart which features a three-armed many fingered woman smiling at an invisible camera.
So that’s what our kids will look like once society rebuilds after global thermonuclear war!!!
I’m sure they will thank us once we explain that the alternative was GPT-5.
Others: unslanted solar panels at ground level in shade under other solar panels, 90-degree water steps (plural), magical mystery staircases and escalator tubes, picture glass that reflects anything it wants to instead of what may actually be in the reflected light path, a whole Background Full Of Ill-Defined Background People because I guess the training set imagery was input at lower pixel density(??), and on stage left we have a group in conversation walking and talking also right on the edge of nowhere in front of them
And that’s all I picked up in about 30-40s of looking
Imagine being the kind of person who thinks this shit is good
But just focus on the vibes. This diverse group of young mutants getting an education in the overgrown ruins of this university.
Not sure how it ties into robo-feyman at all but the vibes
smh they gentrified Genosha
jumping off a roof with an umbrella for a parachute feels tractable
For example, in the case of physics one could imagine working through very high quality course materials together with Feynman
Women of the world: um, about that
@Soyweiser @dgerard From day one he’s been there to nudge, not report.
@Soyweiser @dgerard Everything is clear now!
Since I went looking for links, I thought I’d include them for others wondering.
Thiel invests in Polymarket: https://finance.yahoo.com/news/polymarket-raises-70m-thiel-founders-140000178.html
Nate Silver joins Polymarket as advisor: https://www.prnewswire.com/news-releases/nate-silver-joins-polymarket-ahead-of-2024-election-as-trading-volume-tops-400m-302198424.html
@Soyweiser @dgerard WOW!!! and Silver wants #Biden to drop out. No collusion there??? Thanks for providing context.
@Soyweiser @dgerard it does track with silver going off the deep end in 2016
He started getting high off his own supply and thinking he could do climate science better than climate scientists back in 2012.
https://www.huffpost.com/entry/nate-silver-climate-change_b_1909482
@blakestacey I remember hit talking about Stacy Abrams election for governor and saying tah maybe all the illegal shit led to her loosing but there isn’t good research and so it doesn’t matter. When that happened my thoughts about him changed a great deal
