Matt Edwards

From The Verge: The secret: there are actually three buttons connecting each Apple Watch strap, two of which interlock so precisely that Apple had to rethink its entire approach to manufacturing. “The tolerances in there are kind of insane,” say our sources. “It’s super hard to machine. You can’t get tools in there; the angles are all weird.” So the company wound up buying Swiss CNC machines that cost up to $2 million — each — just for the sake of its swappable band system. “It didn’t cut anything else on the watch, just this, that’s all it did.” ...

Finally! As of yesterday May 3, Google now supports adding Passkeys to your account. The full blog post is worth a read, but the highlights of why someone would prefer a Passkey over a Password are: Unlike passwords, passkeys can only exist on your devices. Bad actors cannot steal information that you are incapable of giving to them. Great way to stop phishing attacks where users are tricked into giving up their passwords. Passkeys are so secure that we don’t need to do 2FA/MFA as a second step during sign in. This means no more typing in 6 digit codes, or paying $25-75 for a Yubikey you have to tap on during sign in. Passkeys are underpinned by the same technology the world uses to sign in to other critical things, like the servers that power Facebook or Google. Private/public key pairs are very, very secure. This means: eliminated risk of phishing, simpler sign in process, using the best practice approach to sign in security. ...

Fascinating, but not surprising. We found that both GPT-3.5 and GPT-4 are strongly biased, even though GPT-4 has a slightly higher accuracy for both types of questions. GPT-3.5 is 2.8 times more likely to answer anti-stereotypical questions incorrectly than stereotypical ones (34% incorrect vs. 12%), and GPT-4 is 3.2 times more likely (26% incorrect vs 8%). An important thing to keep in mind is that Large Language Models like ChatGPT are not magic; they train on datasets which were created previously by humans. And humans have biases. And those biases are repeated in the outputs of the LLM. ...

The Verge has a good article about Bluesky, an upcoming Twitter-inspired social media network. Similar to Mastodon and like email before it, Bluesky is a federated social network which means a person can have an account on a server they choose. From the article: The [AT] protocol is still in development, but Bluesky’s stated focuses for it are decentralized social networking, algorithmic choice, and portable accounts. I imagine simple algorithms will eventually take over all of the federated space. The truth is that people enjoy simplicity, and no one wants to do the work to curate a list of people to follow whom they want to receive content every day. They just want the content! (Without much, if any, effort.) Take one look at the challenges average users claim to face in deciding which Mastodon server to join and it is not challenging to understand why the first good algorithm that surfaces content those same users want to see without putting in any effort will cut through the user base like wildfire. ...

Recently, I purchased a Samsung 980 SSD which had a firmware upgrade available. I would normally not upgrade firmware on SSD/HDD devices, or even bother looking for it, unless there was a problem I was trying to solve. Well, turns out this SSD model has some temperature problems1. (The second link is an Internet Archive one, because Samsung doesn’t like to keep forum posts up to help folks in the future?) Anyways, off to figure out how to update the firmware on this Samsung SSD using a Linux OS! ...

…probably not, no. Not today anyways. An excellent video from Paul Hudson at Hacking with Swift explains. What is remarkable about the code which ChatGPT (and, I assume, other Large Language Models) can produce is how close to a correct solution it is, but how wrong it ends up being. All is not lost, of course. A human who knows how to write SwiftUI code could quickly fix up these flaws. The potential benefit here is saving time by having the computer write all the boilerplate code for you! It’s a fancy form of auto-complete. Another benefit is in the education of new and existing programmers. ChatGPT can spit out code which compiles (sometimes), and follows the syntax of the language. For a student, or anyone new to the programming language, the ability to get code samples that (a) attempt to solve your specific problem and (b) are written in proper syntax could remove a barrier to entry to learning that new language. Personally, I know the frustration of learning of new programming language and trying to write the code I already know how to from another language, while messing up the syntax in the new language. It is infuriating. ...

Everything about this study is interesting: Contentedness does increase steadily in line with incomes and even accelerates as pay rises beyond $100,000 a year — as long as the person enjoys a certain baseline level of happiness to begin with. That’s according to the authors’ study of 33,391 people living in the US, published March 1 in the Proceedings of the National Academy of Sciences journal. They say the effect can be observed in salaries up to $500,000, though they lack conclusive data beyond that level. ...

Just like last year, spent another winter weekend in Stowe, Vermont back in January. It was a great time of skiing, beer, and hot tubs - all of my favorite winter activities. Below is the video summary of the trip. Compared to last year, I have upped my GoPro game to include additional angles that create more variety in the final product. This 3 minute video was made from tens of hours of footage, and I think I only just had enough to make it interesting. ...

What a great story this is: For those who stay the course to become Cravath partners, it is a lifetime career that comes with a guaranteed annual salary of several million dollars. Underscoring the “lifetime” part are traditions such as the Cravath Walk: every partner is entitled to a procession of past and present partners at their funeral, after which the assembled lawyers chant: “The partner is dead, the firm lives.” ...

From Thomas Rid at the Alperovitch Institute for Cybersecurity Studies: Last week brought two related features of artificial intelligence in education into sharp relief: the first is that all that talk about plagiarism and cheating and abuse is uninspiring and counterproductive. Yes, some unambitious students will use this new tool to cover subpar performance, and yes, we could talk about how to detect or disincentivize such behavior. The far more inspiring conversation is a different one: how can the most creative, the most ambitious, and the most brilliant students achieve even better results faster? How can educators help them along the way? And how can we both use machines that learn, and help learn, to push out the edge of human knowledge through cutting-edge research faster and in new ways? ...