10,158 karma · joined December 4, 2013
How magnanimous: they make the effort to support the OS their entire company is based on? It's not just for training the model and hosting them: it's also to run them, where they constantly use Linux to do their thing.
Without Linux, there's no OpenAI.
So I take it the Linux community should be so grateful that OpenAI supports Linux, right?
That is not how it works: the "but it works for me" is a logical fallacy.
Haven't you followed a bit all the posts about models, for example, escaping their environments?
The "but it works for me" is a random data point: an anecdote.
That there are people / companies where models ran wild and destroyed files / messed up projects is a fact.
When there are documented cases of stuff gone wrong, people must find ways to protect their data.
We literally see posts frontpage, daily, about AI sandboxing and we regularly see posts about AI escapes or AI SNAFUs.
And you come and post "show me your prompt, for it works on my machine"...
Most of what is in TFA is correct but it only begins to describe the problem.
My takeaway is very simple: the world runs on broken spreadsheets that are full of bugs and wrong assumptions. And the PowerPoint presentations are shown to business people taking decisions based on the wrong numbers coming from those spreadsheets. And it's the world we live in.
> It was common for the data to be split over two or more sheets.
And when the sheet are separated, for example, by years, that's even more bugs. Stuff is counted twice. Others not at all. The concept of when an entry should be "closed" when it's opened on year X and closed on year "X + 1" is something that bewilders spreadsheets users.
> and errors as the spreadsheets grew or were adapted to new data
Errors in spreadsheets are the big one: spreadsheets are full of errors.
> Another problem with spreadsheets is that the amount of data they can contain is more limited than when you use other data analysis workflows.
Another gigantic issue is that the notion of time is broken in dynamic spreadsheets: not because a spreadsheet cannot be written to correctly deal with it. But because the spreadsheet users don't know how to properly model how values relates to varying time (typically the spreadsheet shall work, for the cell that fetches the value, for the time value of "now" and that's it).
> I know there are lots of people who love using spreadsheets.
It's because you can cosplay being an actual programmer when you're not. The result, sadly, is exactly what you'd expect: buggy spreadsheets full of broken assumptions.
Which are then sent to those meaning real business to be rewritten as dedicated apps...
P.S: as of now I'm working on finance stuff... Same old story: a spreadsheet that has overstayed its welcome, it became gigantic. It's a pure mess of fetching values (and fetching way too many values, which creates technical issues) and bogus little things left and right. So what are we doing, again? Porting that spreadsheet to a proper dedicated app that can, correctly, deal with a proper amount of data, while fixing all the little glitches and gotchas too complicated to fix in a spreadsheet.
The Euro is only 25 years old: it would be pure folly to make the "reserve currency" something that recent.
Then... One country of the eurozone already partially defaulted on its public debt (Greece, in 2015). And France is running an insane deficit: so bad that there are now talks of the International Monetary Fund taking control of France's public finances. France had to raise the yield on its debt to its highest level since nearly 20 years. France cannot reach the "only 5% of GDP" in yearly public deficit, on top of an already insane public debt: it is snowballing and the only outcomes are going to be miserable for the people (and for the EUR).
An economist, before the EUR began circulating, explained by which mechanisms the EUR would lead to Spain, Greece and then France default on their public debts. That economist explained how the EUR would lead to "too many secondary houses in Spain" and "too many public servants in France". When Greece did default on its debt, that economist said: "I was only wrong on the order on which these countries would default".
One would be crazy to make a reserve currency a currency that's a mix up of countries that have different productivity and different fiscal laws.
The EUR is one of the worst currency ever conceived and it could turn out to also be one of the shortest lived currency.
In other words, whether they know it or not, every single bank in the world is involved in money laundering.
P.S: I don't know if that's true or not but some people have made the case that there are, percentage-wise, less transactions linked to crime on the public, transparent, ledgers that are the blockchains than in the real banking system.
Exactly. Not only do I blocklist every known Meta domain at my DNS (I run the unbound DNS server at home) I also took all the IP blocks from Meta's ASNs and have automatically blocked them at the firewall level.
And I did that before doing so was, like now, just one LLM prompt away.
1 billion people are on Telegram: they may not be much better than WhatsApp but Telegram does not have Instagram, Facebook, WhatsApp and is not worth 1.5 trillion.
Really: just fuck the Zuck and his PHP/JavaScript ecosystem. Just get off, entirely off, that crappy abusive ad-ridden, surveillance ecosystem.
Delete your accounts today.
If it comes that, I'll put my real monitor in front of that DRMed monitor (doable with flat screens) and I'll have a private LLM, in real-time, pushing on the real monitor (the one I look at), an ad-free version of the DRM'ed monitor.
DRM is nasty but private AI models are already more powerful and they'll only ever got better. And those open-weights are never ever going back in the box.
At long last we've got very serious contenders actually catching on that have "barely any JavaScript" (TFA's words).
We switched to SSE and couldn't be any happier.
Any tech that allows to reduce the need to reach for JavaScript on the frontend is a godsend.
But both are done. Grading inflation is very real.
The problem is not specific to US uni: in many western EU countries education level is going down the drain at a totally alarming rate. We're talking about kids coming out of primary school and arriving at secondary school with "good grades" although they can barely write, even in their native language.
An official report from the OECD is saying stuff like: "29% of 15-year-old French students failed to reach Level 2 (the baseline level of basic proficiency needed to participate fully in modern society) in mathematics.". We're talking about nearly 1 in 3 people that cannot add and subtract numbers.
And it's not specific to France. It's the same in other western countries. Education in the west is going down the drain while students in Singapore, Japan, China, etc. are hard at work.
People are going to say it's "inhumane" to not grade-inflate. That we should let students pass the first semester without grades. But why stop there? Why not let students who cannot write, cannot add numbers, get their diploma? Neuro-surgeons? Plane pilots? Why not just pick people randomly in the population and assign them a role in society? No more math/physics/chemistry in first year of medicine: anyone can be a doctor.
Really: where do you draw the line? Why stop after that first semester?
It's Idiocracy. And people left and right, just like most in this thread, explaining how all the ideas that let to this state of affair are wonderful ideas that makes lot of sense because school/college/uni is "too difficult psychologically".
The little precious snowflakes, we need to cuddle them.
"adjust the sizes of the available next characters": you mean assigning less bits to characters the most likely to come next, and more bits to the least likely characters to come next?
So it's not totally unlike adaptive Huffman coding but instead of updating the tree on already visited characters, here the tree with the weights of the possible following character is predictive?
When in the end they're just petty thieves and scammers.
Now as basically every Patrick Boyle vid: it's just really good.
Typical Patrick Boyle, talking about the non-innovative 165 pages long "paper" Leopold wrote:
"For scale, Watson and Crick's paper setting out the double-helix structure of DNA, arguably the most important scientific paper of the last 100 years, is two pages long"
Yeah.
When the thing has to verify that an input parameter, passed as an integer, is between 0x000a0000 and 0x000affff and the model decides to pull a library and converter the integer to a string and then do substring matching on "0x000a....", you kinda get an intuitive feel as to how these thing do "reason".
Don't get me wrong: I pay three AI subscriptions and use models daily to help me.
Which is why I know the bottleneck is the shit code the models do produce.
> but I am also wondering if agents would be more useful if they had access to the same infrastructure, software and environments we do as human developers.
I don't know if you're been living in a cave or something: for they do have access to just that.
U.S. M2 money supply tripling in 15 years. US gov public debt more than tripling over the same period.
People from the middle class being unable to buy a house.
The money printer has been going brrrrrr since a very long time and everything got badly distorted.
My bet is that when the next financial crisis hits, we'll see helicopter Ben and hear the brrrrr again.
The crisis in one webpage:
P.S: that's not talking about a country like France, where the IMF shall have to intervene soon as public spendings are totally out of control and the public debt is running wild.
Why is economic growth needed?
Because politicians whored for votes and "kept" their promises by creating ever more public debt, spending more than they made (usually while lining up their pockets and the pockets of their relatives, including their mistresses).
Politicians are so corrupt and so incompetent that none of them would even be accepted as president of a table tennis club.
Capitalism doesn't need endless growth to work. It's governments spending more than what they should that do.
I've dreamed of an AI automatically changing the accent to one that I can understand for a long time.
"Unbearable" doesn't begin to describe that accent.
BTW I don't know what's "whitening" about it: there are plenty of blacks or asians born in either the US or the UK or Australia that speak with a flawless local accent. There's no need to attribute it some racist connotation.
And the reason this is being done is that there are probably many people like me who simply cannot suffer that accent.
P.S: btw you know, it's fine if you can't stand my lovely french accent: no harm taken.
Not just for the war on anonymity. NGOs, often funded by government money (so missing two third of their name), have unified on other topics.
One example that comes to mind is teaming up to organize the biggest influx of migrants into the EU following the arab spring movements: there have been hidden camera recording NGO employees explaining how to dodge controls, how to answer questions (by lying), and they've been actively helping sneaking people illegally in the EU.
A good question one could ask itself before downvoting me is this: if it's a fact that NGOs are teaming up, worldwide, to lie using the rethoric of "child safety" [1]... On which other topics did NGOs lie?
Or are we to believe all NGOs are good and well-meaning, except those working to end anonymity?
And before I forget: in Africa for example it's fully well known that NGO typical means "DGSE":
https://en.wikipedia.org/wiki/Directorate_General_for_Extern...
There are french "doctors" in some of those NGOs who know less about medicine than what I know about medicine by having ex-girlfriends that were doctors.
By squinting one's eye just a little it look like NGO is not that far from a synonym of "shady undercover gov op".
And btw why should any of my taxpayers dollars go to even a single NGO if it's non-governmental?
And another big one: while some of those NGOs have been hard at work passing legislation which as a a result gets people arrested for "hate speech" (or even just for saying "christian prayers in their heads") in the UK, other NGOs are even harder at work making sure that the absolute most horrible of all hate speech cannot ever be criticized "because it's religious".
The double standards are literally insane.
If I was in the UK, I'd GTFO. And not to the EU.
[1] speaking of which, btw, there are serious allegations about NGOs participating in child trafficking, for example in the Haiti SNAFU
Same.
But then vlogs about people sailing can be extremely successful, with millions of subscribers. But then I take it there's more appeal in seeing young couples, with a six-pack dude and a beautiful woman in bikini, sailing around the world than reading description of the various kind of waves.
We don't:
enterprise
entreprise
It's not totally unlike "center", which is "centre" in french and also "centre" in UK english (but pronounced "center", unless I'm mistaken).
Now the problem ATM is that OpenAI, for example, had to cut the price of two of its top 3 models by 80% to counter the chinese models: if you delay your models and a competitors takes over the market, you'll soon be out of bucks and won't be able to rent to Google and Amazon etc. the machine needed to make your new findings.
I know people don't want to hear it but: these companies are running at a loss.
And they're facing competition. Wait until a "good enough" is etched on silicon (by AMD or other) and outputs 70 000 tokens/s: the deal is going to change, once again, once those come out.
The energy, the hardware, the debt, the cost to train, the cost to run, the competition, etc. all have to be taken into account.
P.S: I think you miswrote "diseases"
The project that is full of bugs and not really working?
I probably missed something but I was under the impression that even a "simple" translation like that couldn't be properly done and that the result was, well, buggy?
Where's that thing at?
This way I don't pay tens of VMs "tax" but basically only three (plus one or two VMs I use for testing enhancements to my VMs provisioning / optimization / securing setup, when I work on that).
I don't use LXC (I could) but regular containers, inside VMs.
I'll look into the stripped down "micro" VMs but then I don't spend my days launching VMs/shutting them down so it's not a big deal.
> So rather than pick one, this post advocates layering both, in the spirit of defense in depth: a sandbox VM wraps your containers along with the whole toolchain, and that sandbox reaches the internet but has no route to anything private.
Yup it's the only proper way.
And that is true not just for AI harnesses/agents (that shall try to escape), but also for stuff like Plex/Jellyfin/Immich/private pastebin etc.
If you care about security, there really simply is zero reason to run containers on the bare metal.
Why microVMs? I never ever run a container, AI harness or other, in something else than a full on VM. I could use a microVM but in any case I really don't see why I'd run a container on one of my bare metal OS: the place of a container is inside a VM (or microVM).
Especially for AI harnesses where the threat of an escape is very real: the more defense in depth, the better.
And If I can use rootless Podman instead of "rootfull" Docker, the better. Most of my containers are Podman btw.
> My goto solution for this are apple/containers.
To each his own: my goto solution is an actual server on my LAN with shitload of cores and memory and plenty of scripts to provision VMs etc.
I really don't understand why people are YOLO'ing containers on their bare metal OS.
As a teenager I remember I was amazed at the speed at which Johnny 5 would consume that book.
Is it mostly seen the same way. What is unacceptable is removing the freedom (which you mentioned) of people who prefer to run models locally.
Just as people have the right to tinker at home with DIY and electronics, fully knowing they won't compete with the latest ASML machine, people are free to use open-weight models at home.
BTW Minimax H3 running local can generate amazing short vids quite fast: about 50 seconds to generate a 7 seconds vids on a 4090 (depending on the settings). I've got a friend who spams my Telegram daily with such (no censorship and NSFW btw) vids.
I don't run local but I'll defend the rights of people who want the freedom to do so and I won't look down at them from my high-horse talking about "electricity" and "productivity".
They complain that the models assign too many preeclempsia to black women (I had an ex who was a midwife so yup, I know what it means): what's bad about that?
Anyway, let's consider I missed something and it's actually a "racist" model.
What about women?
> Similarly, consistent exaggeration of the majority gender aligns with previous findings that LLM outputs can skew toward gender stereotypes in health care.
What stereotype? That only biological women do lactate? That only biological women can get pregnant?
To me TFA --and it's not clear why it gets upvotes on HN-- is at approximately the same level as the published study in a serious publication about how dogs from heterosexual men were more likely to rape other dogs than dogs belonging to women or trans.
It was a fake study, made on purpose to show how badly woke academia did become. And it got published, highlighting how crazy woke academia and research papers had become.
From the summary I'm reading here, it sounds like we're deep into woke-signalling territory.
And as a sidenote, I'd say that when the study begins with: "The research – “Evaluating the Potential of Reasoning Large Language Models to Perpetuate Racial and Gender Disease Stereotypes in Health Care”, ", there's a problem already.
Are taxpayers dollars funding this?
Same. Used to do MX (Yamaha 250 2 strokes) when I was below 30 y/o, as well as off-road/dirt biking, but it's on the road that I got one terribly bad accident (flew over a car than ran a red light while I ran an orange one), with an XR 350, where I ended up unconscious and woke up without being able to see anything (freaking scary and I did last about an hour maybe? Don't remember but I woke up in the ambulance and only started seeing while at the hospital).
I also another crash as a passenger on a... Vespa. Insanely dumb crash.
Both times I escaped mostly unarmed (broke a thumb requiring surgery to put some kind of a nail in it and then remove it).
My brother got a terrible tank slapper at the absolute worst moment: usually you can accelerate to lift the front wheel to get rid of it but there were cars in front of him... He literally flew over his own bike (a Triumph Street Triple) without hitting anything: just due to the tank slapper. He thankfully stopped gliding before hitting the cars in front of him (and nearly got hit by his own bike).
I love bikes but I don't do any anymore and they do freaking scare me. As a father now they're actually my nightmare.
My brother bought a new Triumph (once more a Street Triple) and he's got a CBR 600RR for the track but I just quit motorbikes altogether.
> I have cats, they won't care if I eat it, so long as THEY get fed by someone.
Your wife and friends will care. Don't eat it. Glad your wife is OK: could have been worse.
You downvoters really cannot see what's wrong with a Phrack article written in an insufferable LLMish tone that reads just like a 2026 LinkedIn feed?
You are lost.