> Squire works for US Department of Homeland Security Investigations in an elite unit which attempts to identify children appearing in sexual abuse material.
I don't get what the big deal is about mandatory paid vacation. My view is that your total compensation will be set based on the market value of your labor. Some portion of that compensation is given to you in the form of ordinary wages and some portion in the form of paid vacations. If the government mandated paid vacations would it increase many people's total compensation?
In my European mind (I have 25 mandated days off per year), if there was not a mandatory paid vacation limit two things would happen:
1. Further exploit desperate people since those that don't need to work at any cost would steer clear of jobs that have 0 holidays.
2. You would further penalize people with families where both parents work. It is well understood that if your kid is sick you can't really use your sick days and so must use your PTO days. Having 0 available days doesn't play well with having kids (personal experience).
And finally, having mandated PTO allow you to actually take holidays. I heard too many times of companies that offer unlimited PTO and when the employer tries to take some they sabotage him/her or plainly threaten his/her job security.
> total compensation will be set based on the market value of your labor.
No, you do not want that.
The market value of most people's labour is very close to zero.
Left to the market most of the population would live just below starvation, a very small group of owners would live very well, and a small group of artisans would do OK supporting the tiny group.
The easiest answer is yes, since many Americans currently earn minimum wage with no paid vacation, minimum wage with mandatory vacation would be an increase in total compensation. I don't know how paid leave regulations impact wage growth in general, I'm sure there is research on this but I didn't immediately find anything.
Another way to think about it: why do we have building codes? We don't want to incentivize builders to cut corners that would risk an electrical fire or falling down in an earthquake or something in order to offer a cheaper price, so we make it illegal. If unsafe buildings are allowed, it makes it difficult for safe builders to stay in the market. Similarly, we don't want to incentivize workers to sell their labor with zero leave in order to offer a cheaper price, because that risks unhealthy and insular communities (literally unhealthy if people can't take sick leave), poor mental health, unhealthy childcare practices, an unhealthy civic environment if people can't take time off to vote or volunteer, etc. The labor market is competitive and people will sacrifice paid leave if they have to, because they need money to live, so we should make it illegal to remove the incentive.
Unless you have a union, there's a dramatic power imbalance between you (the employee) and the employer at the negotiating table. I'd urge you to read up about the 19th century labor movement and what conditions prompted it.
Wages and time off are not frictionlessly interchangeable in the vast majority of jobs. Mandating minimum levels for both helps make sure people have access to both.
For a lot of us, work is not our life. Turns out that most people really want a paid vacation. Smart Capitalists know that it's easier to extract value from workers with higher morale.
If you would rather trade your paid vacation for an extra week of pay, I am sure you and your boss can work it out. Companies pay out unused vacation all the time. Just don't ruin it for the rest of us!
My experience working at Apple was that private information does not leave the device. All our training data came from contractors who were hired to perform data collection and not from the general public.
If Apple really walked the walk about their privacy marketing, it would sunset all their APIs used for tracking, making impractically hard for advertisers to track you. Not warning "do you still want to be tracked?", but remaking their whole stack to make tracking unreasonably hard.
Currently I see Apple as safer than, say Google or Microsoft, but not as the privacy bastion it claims to be.
It's clear it doesn't bother you, but I'll try to explain my posture.
Years ago, Apple's Weather app sourced their data from The Weather Channel. That meant these three tracking options ragasrding your location:
- Always share - You get real-time weather alerts, very useful some seasons
- Share while using - You get current weather, but lose real-time alerts
- Do not share - Might as well uninstall the app
Then Apple made Apple Weather, which collects weather data from multiple sources, and is supposedly safer to share real-time location with since Apple won't share it with anyone. Before this, The Weather Channel had the real-time location of millions worldwide, and all Apple had for privacy was that prompt.
This is the kind of stack reengineering I'm talking about, that makes privacy a real proposal, but applied deeper so it really makes a difference.
Unless you're some sort of globetrotter going to a new city every week, the app is quite usable just by adding your city.
>Before this, The Weather Channel had the real-time location of millions worldwide
Are you sure apple wasn't proxying the traffic through their servers?
edit: for instance the stocks app very prominently shows the data is from yahoo stocks, but if you check "most contacted domains" in app privacy report, they're all apple domains. It doesn't contact yahoo at all.
>You are one PRISM type request and one gag order from a silent update changinf that.
Wouldn't the bigger issue be that they can abuse the same thing to grab camera and or microphone from your phone? Probably more useful than airpods too, given that a phone's always on, unlike airpods.
People are way misinterpreting OpenAI’s intentions here. The idea is that OpenAI could propose to enter into joint ventures with industry partners where each side gets a share of the rewards generated by the joint activity. This is would only happen in the cases where it’s an attractive proposition to both sides.
From the pharma side, I have heard discussions with other technology companies who insisted on a share of discovery revenue. Nothing has ever killed discussions so quickly.
Same, but it’s hard to imagine these kind of alternative medicine services are driving huge premium increases. Because they don’t work they tend to be pretty cheap to provide.
The number of times I know that my instruction is in context, but it’s forgotten, is countless at this point for me. My experience, both ad a clinical psychologist and developers, is that there is a convergent trend in how I speak to both clients and AI. I can view much of my therapist's approach in how I try to highlight the important things to focus on to achieve progress. Often, it’s about helping the client articulate and understand what’s important to them and how they rank these priorities. The same applies to AI. It feels obvious now that the problem with attention and context is the lack of hierarchy or levels of importance. We know that we have, probably biologically based, three types of memory: short-term, intermediate, and long-term. Long-term memory is what you use with MCP, web search, and RAG. Shorter memory is the current response, and intermediate memory is the current context. When assume this, in my interactions with an agent, it makes perfect sense where they falter and what they forget, in the exact same way as people. It feels more and more like talking to a human, with same weaknesses in logic, reasoning, and focus.
I came here just to complain about that :-) All LLMs I used seem to give more weight to things at the beginning of the context window and omit many details. Eg. I tried this simple thing: pasted a friend's and my CV into Gemini and asked it to recommend topics for a joint conference presentation. Results depended greatly on the order of CVs pasted in.
That's because when they say "long context window" they're lying and they actually mean that they support a long input prompt that is still compressed into a small context window. (Typically by throwing out tokens in the middle.)
An actually large context window is impossible due to how LLM attention works under the hood.
These aren’t really indicative of real world performance. Retrieving a single fact is pretty much the simplest possible task for a long context model. Real world use cases require considering many facts at the same time while ignoring others, all the while avoiding the overall performance degradation that current models seem susceptible to when the context is sufficiently full.
You literally just shift the window over by to the next token once you reach the max amount of tokens you want for context window, NOT with what you train on, (only limited with memory now)
This has obvious issues since you're now losing information from the now unseen tokens which becomes significant if your context window is small in comparision of the answer/question you're looking at. That's why companies try to give stupidly large context windows. The problem is they're not training on the large context window, they're training on something smaller (2048 and above). Due to how attention is setup, you can train on a small amount of context and extrapolate it to any number of tokens possible since they train via ROPE which trains the model because on words and their offset to the neighboring words. This allows us to effectively x2,x3,x10,x100 the amount of tokens we generate vs train with with some form consistency BUT still cause a lot of issues consistency wise since the model approaches more of a "this was trained on snippets but not the entire thing" situation where it has a notion of the context but not fundamentally the entire combined context
That’s a very basic way to keep the LLM inferring past the context window size (there’s better, smarter ways) but that’s not at all what the question was which is how they train a 2M token length window. My understanding at a basic level is that you need corpuses that are >2M in length for training data which is where the problem comes in for - there’s only so much long form content and it’s swamped by all the smaller stuff. I think there’s probably tricks now but I suspect it’s still largely an open problem.
AFAIK nobody does that. They train on much much shorter text but with use tricks in the position encoding steps that can be extrapolated by the LLMs. Lile ROPE and YARN etc.
AFAIK (not much) it definitely helps to train on longer sequences even with rope/yarn and is needed if you care about long context performance (and not just the long context capability).
It's not the most energy efficient workflow, but I work on relatively small codebases and I made a tool that let's me dump all of it in an LLM with a single copy/paste. This works surprisingly well with Gemini 2.5 Pro (1.000.000 ctx).
The only real mistakes it makes are some model specific quirks, like occasionally stripping out certain array index operators. Other than that, it works fine with 150.000 token size conversations. I've gone up to 500.000 with no real issues besides a bit of a slowdown. It's also great for log analysis, which I have maximized to 900.000 tokens.
Most attention implementations can work across an arbitrarily long context.
The limiting factors are typically:
1. Often there are latency/throughput requirements for model serving which become challenging to fulfill at a certain context length.
2. The model has to be _trained_ to use the desired context length, and training becomes prohibitively expensive at larger contexts.
(2) is even a big enough problem that some popular open source models that claim to support large context lengths in fact are trained on smaller ones and use "context length extension" hacks like YaRN to trick the model into working on longer contexts at inference time.
The model will use the full context if it's been designed well, but you can still increase the size of the window on models where it hasn't. It's just pointless. People who don't know much about LLMs will still think "bigger number is better" though.
If a model is not making use of the whole context window - shouldn't that be very noticeable when the prompt is code?
For example when querying a model to refactor a piece of code - would that really work if it forgets about one part of the code while it refactors another part?
I concatenate a lot of code files into a single prompt multiple times a day and ask LLMs to refactor them, implement features or review the code.
So far, I never had the impression that filling the context window with a lot of code causes problems.
I also use very long lists of instructions on code style on top of my prompts. And the LLMs seem to be able to follow all of them just fine.
You wouldn't ask a human to do that, why would you ask an LLM to? I guess it's a way to test them, but it feels like the world record for backwards running: interesting, maybe, but not a good way to measure, like, anything about the individual involved.
I’m starting to find it unreasonably funny how people always want language models to multiply numbers for some reason. Every god damn time. In every single HN thread. I think my sanity might be giving out.
Since grok 4 fast got this answer correct so quickly, I decided to test more.
Tested this on the new hidden model of ChatGPT called Polaris Alpha: Answer: $20,192,642.460942336$
Current gpt-5 medium reasoning says: After confirming my calculations, the final product (P) should be (20,192,642.460942336)
Claude Sonnet 4.5 says: “29,596,175.95
or roughly 29.6 million”
Claude haiku 4.5 says: ≈20,185,903
GLM 4.6 says: 20,171,523.725593136
I’m going to try out Grok 4 fast on some coding tasks at this point to see if it can create functions properly. Design help is still best on GPT-5 at this exact moment.
You got the right idea there. They wouldn't actually show up in your Fidelity account but there would be a different website where you can log in and see your shares. You wouldn't be able to sell them or transfer them anywhere unless the company arranges a sale and invites you to participate in it.
When the commission is to create the most impressive structure possible, anything less would be a failure. That is just how the Catholic church rolls. See most European art and culture for the last 1500 years for details.
The Cologne cathedral took over 600 years to finish because the original plans got lost along the way. it was paused after 300 years! For the following centuries, many generations only saw the same unfinished state with the crane on top.
Most cathedrals and monuments are like that because until recently in human history, they took a long time to build and so the original architect would die, the financing might collapse, etc. Heck, this happened to Gaudi; the remarkable thing here is that the people after Gaudi wanted to continue his vision as much as they could.
The Washington Monument in DC, for example, famously is different colors because they had to change the source of marble during construction when funding halted for a time.
Fiction, but you if wonder about things like this, you might be interested in The Pillars Of The Earth series about the building of a cathedral in 12th century England.
Not sure “celebrities” were such a thing as they are today. 7 centuries was before the reformation and things were pretty austere. Surely nobles celebrated things and there were favored artisans but celebrated as crassly as we do today in such abundance. I don’t think the media existed to allow that to take place.
Eh, they certainly weren't celebrities in the same way, that would only be possible with modern broadcast media. But people like the pope, kings, and dukes would be pretty close. I would expect the average medieval peasant would know who the pope (or popes, depending on the date) were, and at the least who their king was, as well as the relevant nobles for their village. And I wouldn't be surprised if they knew who the neighboring kings and nobles were. A peasant from the Iberian peninsula might not know who the king of Poland was, but they would likely know who the French king was and likely who was emperor of the Holy Roman Empire.
And medieval people definitely built monuments to themselves. A great example is Battle Abbey [0]. The official reason it was built was as penance for William the conqueror killing so many English, but there is definitely a strong case to be made that building such a grand abbey was in 0art to signify the new Norman rule and to remind people of who was in charge. They weren't venerating the architect, but it was very clear to everyone who paid for the abbey and William remained very much linked to the structure. That would have been one of the most impressive buildings for a very large area, even it's ruins remain impressive nearly a millennium later. It's a religious building, but it was even at the time very much linked to a secular ruler (inasmuch as the rulers of the time were secular).
It really depends on what you mean with "know" here.
The legend say that when the king tried to flee the revolution he was only recognized due to a coin with its face engraved in it. A teacher taught me this one with a variant where the king itself gave the coin to pay in a tavern. Now even it is just a legend, that also gives an interesting reflection on what it means to be famous at this time.
A typical Iberian peasant probably wouldn't have heard of Poland. The King or Emperor would be "the King" or "the Emperor" and might as well live on the Moon.
Not many people realise that the more distant locations in Shakespeare's plays were close to science fiction. If you were a British peasant visiting "Verona" or "Venice" was like visiting the ISS. You might get swept up to fight in France, and there was a tiny chance of joining the navy. But most people spent most of their lives within a tiny area, with little idea of what was happening elsewhere.
So cathedrals were stunning. If you somehow visited a cathedral city you'd be struck dumb by the size - unimaginable to someone who grew up on a small holding.
reply