I think you're underestimating the "more expensive". It's not a 30% discount or similar. API is at least 10 times more expensive (own experience), ~40x as expensive if [1] is accurate.
At that price difference, lots of things that are viable to do in whatever harness the subscription allows are simply impractical via API, you'd be spending a whole developer's salary just for the benefit of using a different harness.
I tried for example Deepseek V4 Pro for a bit (via API pricing), but it's _more expensive_ than Sol via subscription, even though $1.7/Mtok is a lot less than 20$/Mtok.
> API is at least 10 times more expensive (own experience),
More. On CC I'm currently burning ~$5k a day (according to claudes own metrics) and I use up the weekly quota in about 3-4 days. Meaning I'm eating $20k worth of tokens for a $200 sub.
Current session I have running in the background, spinning up tests mostly
> At that price difference, lots of things that are viable to do in whatever harness the subscription allows are simply impractical via API, you'd be spending a whole developer's salary just for the benefit of using a different harness.
Could you elaborate on that?
Off topic:
You did the ISBN visualization thing! Awesome!
Which OS are you using? systemd-oomd (or similar) afaik is enabled in many modern releases and fixes it. The default settings (on Arch) make it kill stuff when your CPU is spending 60% of time on reclaiming memory over some time period. So you'll get 20 seconds of lag, then it is fixed.
By killing a program, yes, but how else do you expect it to work? If your RAM is not enough you have to kill stuff. And I really doubt you mean to say that Windows is _more RAM efficient_ than Linux, because that's kind of absurd. Maybe Firefox is less RAM efficient than Chrome.
Edit: MacOS I wouldn't doubt being better in RAM management than the others because they have been selling their devices with ridiculously low amounts of RAM for decades.
How does it work on Windows and MacOS where I don't get the same freezes and random programs aren't killed by the OS? Because I'd expect it to work like that.
There is this false dichotomy, a black and white kind of thinking on display.
> And I really doubt you mean to say that Windows is _more RAM efficient_ than Linux
What I'm saying ought to be clear to anyone willing to listen. My system should not get into an unrecoverable freezing state when I am doing routine tasks on a fresh install of an OS. And the community that is supposed to champion this OS should not then shift the blame to me if I complain about it.
"My system should not get into an unrecoverable freezing state[..]"
There is a good precedent for what you want: Safety critical embedded systems. Your car ECU should not be in an unrecoverable freezing state when you want to break.
These embedded systems also are known to be terribly memory inefficient because every bit of memory is statically assigned and never reused.
So, it is not that what you want is impossible, it's just wasteful and expensive.
Linux and Windows sit on slightly different points on this spectrum (opposite of said embedded systems) and what you perceive as Windows' better behavior under memory pressure has a price too.
and killing programs and/or freezing the OS is somehow more efficient? By what metric? Certainly not by any UX or productivity one.
I've moved to Linux from Windows some years ago but the first thing that shocked me was precisely the abysmal performance around low memory availability.
Got my system stabilized eventually by means of some magical incantations but this wasn't a pleasant experience and I wouldn't want to repeat it.
It's beyond me how come Linux is so decrepid in this particular area compared to Windows.
Yeah, Windows file system backend is highly object oriented and has a bunch of event driven hooks. Which is great for supporting tools like antivirus. But terrible for lots of small random writes (on area Linux and most UNIX systems excel at).
macOS shows a modal with a list of offending applications and how much RAM they're using. Some of those applications are "paused", marked in red, some are "not responding", and some are responding normally.
You can pick applications from that list and choose between "Resume" (for "paused" applications) or "Force Quit".
Freeze as many applications as necessary to reserve room for the DE so I can choose what to do?
Surely that's obvious to other people aside from me. It's an attended system, not a headless server in a rack hundreds of miles away that must not ever alert someone.
The ideal would be to go into swap without becoming unresponsive, so that the user can quit a program of their choice to free memory. If you really want to make an issue out of it, do something like pre-X macOS, where it would provide a warning when you are getting close to running out of memory, so that a user might have the chance to do something before the current process starts getting swapped out.
> The ideal would be to go into swap without becoming unresponsive, so that the user can quit a program of their choice to free memory.
Linux already does this. The issue is when your swap is full too, or you’ve exhausted your RAM at such a rapid pace that the disk IO cannot keep up with the read and writes to swap.
I’ve seen Windows choke under the same conditions too.
> do something like pre-X macOS, where it would provide a warning when you are getting close to running out of memory, so that a user might have the chance to do something before the current process starts getting swapped out.
This is where the desktop and the kernel being entirely separate organisations (not even projects, but literal organisations) makes things a touch harder. And those DEs aren’t even Linux specific either. So any tooling would have to portable (which, in fairness, should be possible with POSIX in this particular scenario).
I think KDE does provide tools here for gracefully handling memory constraints. But that’s just one desktop environment of many.
>> do something like pre-X macOS, where it would provide a warning when you are getting close to running out of memory
And I've run out of memory while compiling packages (I'm looking at you, composable-kernel) on Gentoo. It's not like it can notify me and I close things in the milliseconds it takes for a few more GCC processes to spawn and allocate and fault all the remaining memory.
Linux can, and does, prevent a program to claim any new memory. Like literally malloc() returns null. How program reacts to that is up to the program.
Also I know linux has `cgroup_enable=memory` setting in /etc/default/grub, that's for Docker/K8S to be able to watch over containers RAM. It's not enabled by default as it slows down OS a little. Don't know if it's related to my first point.
An encrypted hard drive is EXACTLY uniformly distributed random bytes if you do not know the encryption key. No one would be able to tell the difference between a drive that is just purely random numbers or is actually filled with content. (Of course excluding the usually intentionally added readable header)
If this was not the case, the encryption would be broken, and most everyone agrees that good encryption does exist.
The "quality of the probability calculations" as you put it is 100% in this case and any less would be a huge deal (as in - breaks all of the internet).
So, now you just take those same random bytes and use them as the seed for your LLM token choices. The output has the _cryptographically_ proven exact same quality as if you were using a true RNG (which you likely weren't using anyways).
You just need to know your LLM distribution and the encryption key, then with each new token you exponentially increase the chance of knowing whether it fits your encryption key. Without actually affecting the token choice in a perceivable manner.
> choosing different words that it otherwise would
The "that is otherwise would" is carrying all the weight here. "Otherwise" is sampling from a distribution. You just sample from the same distribution but with a cryptographically secure, seeded RNG. https://en.wikipedia.org/wiki/Cryptographically_secure_pseud...
"Knowing the LLM distribution" seems to me like the only hard part because you don't know the context of any random snippet.
>You just need to know your LLM distribution and the encryption key, then with each new token you exponentially increase the chance of knowing whether it fits your encryption key. Without actually affecting the token choice in a perceivable manner.
I struggle to understand the relevance of that comment.
The blue/green token list biasing process literally does cause different tokens to be occasionally chosen. Not only that, but because a different token was chosen at one point, this changes the probabilities of all subsequent tokens, and the resulting later token stream every time it happens. If you had access to the token stream as it would have been, and the watermarked one by the end of the text they will be very noticeably different.
Well I was assuming your comment was saying it's _impossible_ to do this without affecting output, so I was just describing an option on how it IS possible. I saw that token list biasing post and yes that looks pretty bad. I didn't look in enough detail to see whether it's actually clear what Anthropic and SynthID really do. I would assume though their approach is of the smart kind.
The EU told them they can not by default prefer their own maps product when linking from their search product, they have to allow the user to choose.
They could have simply added a selector when a user first clicks on the maps preview in the search result, and then remembered it on device or across that user's account.
But of course, then the user could choose a competitor's product and Google would have to honor it. That would be horrible, so instead they just made the user experience worse for everyone and elegantly made people blame the EU.
I wonder if the EU will someday figure out a patch for this "Deteriorate everyone's user experience in a way that is technically compliant and blame the EU" behavioural exploit...
I'm blaming Google for this, but most people just don't know why it happens. Probably they'll end up mandating that Google must show a notice with the reason and it will be implemented in the worst possible way.
All of these hissy fits Google and Apple are throwing are just making me wanting to use them less and less and I'm doing my best to explain how tech companies are doing malicious compliance to others.
What does it has to do with the topic? And saying “governments do not build” is a crazy antihistorical take: many things, from the space industry to the telco one were built by governments, as well as the energy and transport sectors in many countries.
For the past 50-ish years the West convinced itself that government should not build, and as a result we're suffering from a slow collapse of basic infrastructures. Meanwhile, in China and the Gulf, government kept building and the world's center of gravity steadily keeps moving eastwards.
Yeah, it's very overpowered and should be nerfed - they still haven't patched the cookie consent dialogs out.
To expand upon the silly comment a bit, not sure about this particular initiative, but to me it very much feels like something that should be configurable at browser level, instead of half-assed dark-pattern-filled JS dialogs: https://killthecookiebanner.eu/
It definitely should be, but this just goes to show that regulations are necessary if we don't want companies to unnecessarily screw with users. It would have been better that way for every user, but it would be worse for people who want to sell everyone's personal information.
Oh, and please ignore that this surveillance apparatus is now being used to expell people, all of which already has resulted in many missing people, and even some deaths. Any historical parallels are entirely coincidental.
It's fine + obligation to fix, and daily fine until fixed, and higher fine if you do it again.
I know that sounds like it comes from "it will never work", but that's how active directory, office file format, etc ... got opened, and in a case more similar to here that's how windows n and decoupling of media and internet component from windows internals happened.
The issue is, yes, it takes time.
Another issue, purely PR, is that yes the EU gets attacked repeatedly by people trying to match "respect the law of the market or leave" with "the EU tax US companies because they can't compete".
> It's fine + obligation to fix, and daily fine until fixed, and higher fine if you do it again.
The important proviso is that the penalties need to (on some non-geological timescale) get high enough that the offender is no longer capable of operating. That could mean the fines reach $500 billion, or it could mean the company is barred from operating in the EU, or it could mean people get arrested and assets are seized. But unless the penalties become crippling, it won't matter. It needs to reach a point where the downsides of noncompliance are actually greater than the benefits.
NIS2 allows for the arrest of managers in case of cybersecurity incidents resulting from negligence. I think it's a step in the right direction, but we'll have to see how it plays out.
The solution is to mandate that websites are built for browsers/devices as much as they are for humans. That way you never see a cookie banner because you've set up your device to never accept marketing cookies; A Google Maps link clicks through to Google Maps because you told your browser Google Maps was your preferred online mapping tool.
Why should be? I think it is abundantly clear by now that market competition alone will not fix this. Though that does not stop people from claiming "Yes that failed and people suffered/died, but that was not real capitalism! If only we had a real free market, it would have been better!"
The specific problem with Maps is that Google _can't_ deep link to other providers except by lat-lng or address, both of which throw away the actual context of the PoI in the common case of searching for PoIs. Well, it could perhaps deep-link to OSM-based maps, since their map data is open... but Google, as a proprietary map maker, probably treats OSM data as radioactive.
They could make available a URI using the geo: scheme, but the hacky ways of including a POI as part of a geo: URI (ie using a query string) feel generally unsatisfactory.
Google measures latency at microseconds.100 ms ≈ 0.2% fewer searches. Google tracks this very closely and invests a lot as that's the operational life blood of the company. You can make more money by simply making search easier.
To save "it's easy" and Google chooses to nuke their own products for... I don't know what, stick it the European regulators, strikes me as incorrect. It's because it's incredibly hard to navigate this arbitrary punative regulatory framework.
No one would use the other maps because other maps are worse. As much as euro regulators hem and haw about lack of competition, it's precisely these types of dictats that make Europe not have any viable competitors in tech.
I often hate how loaded/incendiary Ed Zitron's tone is and don't like some of his takes, but I wouldn't call it "some random blog". I run a random blog, that guy's closer to being a proper journalist: https://en.wikipedia.org/wiki/Ed_Zitron
Referencing a blog is fine. But when you're on a forum discussing ideas just forwarding a blog link is bad taste. Summarize the key arguments. Do some work.
It's like if someone is arguing free markets and I just provide a link to a WSJ editorial. It's rude, requires no work on the sender and imposes work on the receiver to read and synthesize and respond to the argument. I see this occasionally where someone will just have an LLM generate a response (which it will happily do for any topic taking any position) and just passing it off.
It's malicious compliance with the intent of having the users misunderstand what's happening (that you don't have to do that but do it on purpose, and instead blame the law protecting them).
And eh, it works for the cookie banner so why change the strategy ? The comment you're answering to prove it still works. As long as people can't be bothered to think and inform themselves for a second about it, they won't stop. Same with the purposefully annoying and unclear and user hostile 'gdpr popup' about third party consent.
I'm not saying I agree with it. I'm saying from Google POV that's the correct move given their intent, the issue then become people like parent.
Yes and no, it's true now but the EU fines are not just "you did that" but "you did that we're going to assume you're going to be better and if you do again next time it will really hurt". Which I think is a fair point of view.
You might still answer too little to late, but given that we're at the starting point to a massive sovereignty push here thanks to the current US admin being proved as a possibility rather than a novelty, I think this matters a lot.
Well no, the cookie banner is a well intentioned but flawed and poorly designed law. Any law that is that technically specific but relies on people without an understanding of how it works is doomed to repeat the same fate. The eu should have seen prop 65 and not sleepwalked back into it.
1. The EU works on intent of the law rather than specific wording and precedence.
2. The law is a very simple "if you want to do it, you have to make sure they know", intended to force information without creating excessive administrative / legal / tech burden, please inform me about what other way you would present it that would reach the goal without being subject to malicious compliance ?
I understand. I'm generally a proponent of the spirit of the law rather than the wording of the law, and I've argued that is how the EU works here in the past. I think GDPR is a much, much better attempt at solving the problem, and has made meaningful change in tech industries. The ePrivacy directive just added pseudo-mandatory popups to every company without a technical lawyer.
Why? The whole point of the GPDR was to prevent medical information being used for insurance and all sorts of purposes.
Then come the lists of what exceptions are approved. Your medical info is used for divorces (anything involving court cases, anything involving criminal law), the police has access to it, your mayor has access to it, tax departments have access to it (think you can not pay tax and pay for your kid's cancer treatment instead? In Europe, think again). Insurance (if you get treatment for getting hurt in traffic your car insurance goes up). Unemployment (if you get treated for anything drug-related ...). Hospitals and doctors can use your medical information without your permission (for billing, for other treatments, for deciding if you should be interned, ...). And so on and so forth.
Oh and there are even silent exceptions. You see, YOU can't sue anyone under the GPDR. You can only ask a specific "supervisory authority" (you can't even choose which one)
They are under control of the executive, and so it is in most cases the currently elected party that decides if your GPDR complaint does anything, NOT the courts. Not the police. Not the public prosecutor. None of that. And it's even closed on the back end: you don't agree with these "supervisory authority"'s actions? Doesn't matter if you're complainant or defendant. You can't sue them either. You can't get a judge on your case, only appointed politicians.
There are even organizations that the GPDR supposedly applies to that have their own supervisory authority. Interpol violated your rights? No worries, file your complaint here in this building. You know, the building with "Interpol" on it in big letters.
So really, we do not even know the full list of exceptions.
More generally, the GPDR was supposed to prevent further encroachment of all sorts of organizations on privacy, with a big focus on medical data. It has achieved the opposite of that. FOR NOW (and not in every country) the only way to get a private medical file is to only use private medical care. For now that is still possible.
It's like the DMA (Digital Markets Act). Prevents organizations from using control of the OS to implement policy. There's a few exceptions though. Google gets an exception. Apple gets an exception. Through specific deals made with these organizations and the EU commissioner of
Nobody seems to have thought to scream into the commissions face: "THEN WHAT'S THE POINT?".
Well, who made those deals? Thierry Breton. He currently serves as a remunerated member of Bank of America’s Global Advisory Council (who have huge investments in Alphabet and Apple).
Yeah, I get why you want to focus on the intent only and not on what practically happened. Theory and practice are very, very different and the EU is incredibly pro-business and uses their power to literally grant billionaires exceptions to laws. That's how Goldman Sachs got it's first communist president (Barosso, who saved Goldman Sachs as president of the EU commission). That's reality, but of course the intent is thoroughly disguised, and you don't want to talk about the difference.
Gdpr whole point was about insurance? Can you please diversify your news source and educate yourself? I didn't even bother to read your pamphlet of a comment after seeing such an obviously wrong first sentence
No it was about privacy. Specifically given as an example in the actual law, privacy of medical data FROM insurance companies. But of course privacy from everyone.
The goal was not even remotely achieved, and this was 100% on purpose.
Then again, the very same people complain about GDPR not being technically super specific.
At some point we have to accept the pattern. People with ideological objections against any limits at all will try to frame any regulation as stupid, regardless of what is in it.
Google really sucks at giving users a choice. The Gmail app on iOS has a browser selector dialog, which asks if you want to open a link in an email in Chrome or Safari. Despite the fact that you can change your default browser in iOS settings if you really want to use Chrome. And even if you choose Safari and don’t ask it to remind you every time, it will still prompt you again later if you’d like to switch to Chrome this time. Couldn’t I just pick Safari and you respect that choice until I say otherwise?
I do wonder if people using Chrome on iOS face the same issue, or if Google respects your browser of choice a bit more if it’s their browser.
At that price difference, lots of things that are viable to do in whatever harness the subscription allows are simply impractical via API, you'd be spending a whole developer's salary just for the benefit of using a different harness.
I tried for example Deepseek V4 Pro for a bit (via API pricing), but it's _more expensive_ than Sol via subscription, even though $1.7/Mtok is a lot less than 20$/Mtok.
[1]: https://tibotattle.com/