vletal

> Messages uses on-device machine learning to analyze image attachments and determine if a photo appears to contain nudity. The feature is designed so that Apple doesn’t get access to the photos.

LGTM

Maybe for you, I am dyslexic and my writing sucks. Even in my mother tongue, let alone English. If I need to give a good impression, I rely heavily on Grammarly because it can reformulate sentences in a way I never could.

Yet for sure, I do not use it for anything sensitive or for work purposes. I do not really trust them.

Yeah. Great mind teaser. I can imagine an imaginary number (-/+) "inf" for divisions by number approaching zero from left/right, yet the algebra would not be possible to define properly because you can approach zero with a different rate.

The undefined cases, where the left/right limits are not equal coulf get imaginary number "shrug" because it would be even less useful.

Of is anyone able to define a useful algebra for these? I'm really curious.

I guess like plenty of bounded processes in nature, this is likely a sigmoid-ish function where close to zero you can approximate the function linearly. Yet yeah, there is some error.

> This is the sort of whoops, sorry accidents ML will make.

Vision based ML. Is this not one of the many cases where LiDAR would have prevented this from happening?

The issue with comparing these numbers is that the distribution of quality and safety of roads is not the the same for data collected by Tesla vs. "human drivers". It is easy to claim that in relatively safe environment of a interstates and highways Teslas did better than human drivers everywhere else per 1M km.

TL;DR; even if this is truly your only metric to judge these systems, you can not simply compare these numbers.

The only claim which I am making is that the p(word|swiped pattern) is inferior to p(word|swiped pattern, previous words).

Multi lingual swipe works just fine on my iPhone with Google keyboard. If it is not it's just because I get too comfy and lazy, not actually "spelling" the words properly. With the dawn of on device language models I believe the typing experience will get even better in the upcoming years.

A link which attempts to open WA via their API and send a message to the bot. Very nasty.

You should submit a link to a site describing the chatbot rather than almost phish users.

You should have seen the raised eyebrows of my psychologist when I told her I do not have a pre-nup.

Yet she might be biased. Most people she meets needed it.

Is the amount of spike protein in the body during an actual infection larger? I.e. is it not still better to get vaccinatet rather than infected?

You were in touch with them since 2021. Did they manage to fix it for M2? Or is it also valnerable?

I wonder how approachable it would be to optimise a custom model for ANE. According to the code examples at the bottom, the current implementation seems to be a custom model, so no generic solution.

Anyway, it seems that we are at dawn of deploying mode cool models which formerly required cloud computation to the hands of the users. Really cool!

Are we going to see more federated learning being pushed to user devices or is it a dead branch only useful for a few use cases?

I really like the debate around this. Everything from "humans are just organic machines" all the way to "we are really special" climaxing at panpsychism.

I am personally on the "we are not that special" part of the spectrum. I wander, what about the rest of HN crowd?

I am not really sure whether overlapping windows was the thing I was missing in iPadOS. I feel like this is gonna fall into the "I will get used to it" category.

Czech TV has a rather useful app which sends notifications about critical events, such as crises, election results, major government decisions...

Once they accidentally sent a rubbish message with a bunch of non alphanumeric characters and a curse word right in the middle of it.

There is a brainfuck variant supporting syscalls. Seems like a great candidate for a llvm backend. You could have mostly any app you want.

> Even if I refuse to interview people as part of my job, will I really thrive in there? I'm certain this process permeates in one form or another throughout the company culture in various internal management, decisionmaking and promotion processes, but that's a separate discussion.

Are you sure you want to work at a FAANG company? You sure you would be happy? What motivates you to try so hard?

In the blogpost they give an example of exactly that

   $ ssh-add -L
   sk-ssh-ed25519@openssh.com AAAAGnNrLXNzaC1lZDI1NTE5QG9wZW5zc2guY29tAAAAIKgGePSwpBuHUhrFCRLch9Usqi7L0fKtgTRnh6F/R+ruAAAABHNzaDo= cadey@shachi

Seems like the key is exposed as ssh agent.

Yeah. I might be generalizing a bit, even though I made sure to emphasize that it is my isolated experience.

On the other hand I felt like a sample of 8 different interviewers from different offices and backgrounds felt representative. If it is true that the interviews are overwhelmingly negatively received, what is the probability that I would get more than half of reasonable interviews? Let alone 7/8.

Well. I definitely underperformed at Meta. Google went great IMO with a few stumbles here and there. No feedback yet. Wish me luck.

Interesting. Might have been just lucky.

I might not have been clear. It's not an act. You are not usually guiding your peer trough your thought process our loud step by step. That's inefficient. Yet that is what you should do to maximize the amount of information passed onto the interviewer. It's just a different setup to which most people need to adjust to.

Interviewing is assessing how a person would perform, not an actual day to day collaboration. There is a gap. You do not talk like you would to a colleague. You have a limited time and it's more one-sided to even the playing field. It's worth preparing for to overcome the gap.

It's like a dance. Your are supposed to lead deriving the solution, interviewer is supposed to follow closely and judge your performance.

If you expect to ace an interview just like that then you also need to be lucky not accidentally getting asked to stuff, which you always need to Google an answer to, basic algorithms and patterns, which you usually kind of know but could slow you down during the precious 40 minutes you have to show off your skills.

And lastly, people do prepare for these. If you do not do that, you might simply make a slightly worse impression than someone who did.

That being said, I mostly did not prepare for the coding interviews. I focused on the design and behavioral interviews.

Not sure what is wrong with preparing for a challenging interview. What is the point you are making?

I've just interviewed for Google and Meta. In total I went through 8 individual coding interviews.

With an exception of one, which was a bit too much like a puzzle, I think the assignments were completely fair examples of what I might experience on daily basis. Not at all tailored to people doing competition style programming, which was what I expected after being exposed to HN for years.

I have to say I enjoyed both preparing for the interview as well as the interviews themselves. And I believe that overall the interviewers were able to grasp how I would code.

What about battery life on iPhones? Is the hardware utilised efficiently?

Given our history of creating solutions to problems and dealing with side-effects later, I'd love to first see plenty of research explaining how is this not gonna go out of hand in a few years.

It's like online voting. It works as long as the government does not have a malicious intent of staying in power indefinitely.

Then the secret word list might include stuff like "free speech" or "Uighur" and the world would go to hell.

Are you American? I feel like you are assuming taking extreme measures to solve US problems. Have you considered how similar measures already affect markets in other countries?

ad. "not a market" - When you do that are you going to also forgive my crazy large mortgage?

ad. "protect tenants" - In Czechia we protect the tenants more than landlords. If you mean that this should scare people from renting that works quite well. People do not have much leverage here to kick out tenants who are problematic or skip payments.