By Jason Snell

January 7, 2025 10:25 AM PT

LLMs aren’t always bad at writing news headlines

On Monday I complained about Apple’s response to Apple Intelligence making mistaken summaries of news headlines. But here’s the funny thing: large-language models are actually pretty good at writing news headlines.

The problem with Apple’s approach is that it’s summarizing a headline, which is itself a summary of an article written by a human being. As someone who has written and rewritten thousands of headlines, I can reveal that human headline writers are flawed, some headlines are just not very good, and that external forces can lead to very bad headlines becoming the standard.

Specifically, clickbait headlines are very bad, and an entire generation of headline writers has been trained to generate teaser headlines that purposefully withhold information in order to get that click. (Writing a push notification is not the same as writing a headline, but it’s at least similar.)

Not to get all Fred Jones on you, but: I was trained to write headlines that made you want to read more, but didn’t withhold information. The idea was to compete for your attention, not pose riddles that could only be decoded by reading the story in question.

The now-dead news app Artifact built a killer feature that rewrote clickbait headlines on demand. It used the complete content of the news story to write a new headline that was always better than the ones being served up by various news organizations. Yes, when given enough information and asked to generate a headline, it turns out that AI is pretty good at the job!

Still, as Mike Krieger of Artifact wrote about in a Medium post, the company also built in some layers of human validation:

When we rewrite a title given a user request, that new title is initially only visible to the user. If enough people request a rewrite of a given title, the rewritten title will be escalated for human review and if it looks good, we’ll promote it the new default title for all users.

One of the principles we use when applying AI to features inside Artifact is to be as transparent as possible. For both Summarize and the clickbait title rewriter, we’ve used consistent iconography (the star-like shape) to denote that there’s AI involvement.

It didn’t work out for Artifact, but there’s something here. Summarizing summaries isn’t working out for Apple, but more broadly I think there’s something to the idea of presenting AI-written headlines and summaries in order to provide utility to the user. As having an LLM running all the time on our devices becomes commonplace, I would love to see RSS readers (for example) that are capable of rewriting bad headlines and creating solid summaries. The key—as Artifact learned—is to build guardrails and always make it clear that the content is being generated by an LLM, not a human.

By Jason Snell

January 6, 2025 4:00 PM PT

Apple Intelligence summaries might get warning labels. That’s not enough.

The BBC, following up on two reports of Apple Intelligence summaries that transformed its own headlines into factually inaccurate text, got a public response from Apple:

Apple has said it will update, rather than pause, a new artificial intelligence (AI) feature that has generated inaccurate news alerts on its latest iPhones.

The company, in its first acknowledgement of the concerns, on Monday said it was working on a software change to “further clarify” when the notifications are summaries that have been generated by the Apple Intelligence system.

Here’s what the BBC reports Apple’s statement is:

Apple Intelligence features are in beta and we are continuously making improvements with the help of user feedback. A software update in the coming weeks will further clarify when the text being displayed is summarization provided by Apple Intelligence. We encourage users to report a concern if they view an unexpected notification summary.

The statement uses the beta tag it has placed on Apple Intelligence features as a shield, while promising to add a warning label to AI-generated summaries in the future. It’s hard to accept “it’s in beta” as an excuse when the features have shipped in non-beta software releases that are heavily marketed to the public as selling points of Apple’s latest hardware. Adding a warning label also does not change the fact that Apple has released a feature that at its core consumes information and replaces it with misinformation at a troubling rate.

Apple is shipping these AI-based features rapidly, and marketing them heavily, because it fears that its competitors so far out in front that it’s a potentially existential issue. But several of these features simply aren’t up to Apple’s quality standards, and I worry that we’ve all become so inured to AI hallucinations and screw-ups that we’re willing to accept them.

We shouldn’t be. Apple’s shipping a feature that frequently rewrites headlines to be wrong. That’s a failure, and it shouldn’t be shrugged off as being the nature of OS features in the 2020s.

So what can Apple do now? A non-apology and the promise of a warning label isn’t enough. The company should either give all apps the option of opting out of AI summaries, or offer an opt-out to the developers of specific classes of apps (like news apps). Next, it should probably build separate pathways for notifications of related content (a bunch of emails or chat messages in a thread) versus unrelated content (BBC headlines, podcast episode descriptions) and change how the unrelated content is summarized. Perhaps a little further down the road, news notifications should be summarized based on the full text of the news article, rather than generating a secondhand machine summary of a story already summarized by a human headline writer.

Beta software contains an implicit promise that the developer will actively work to squash bugs and make the product better before it goes final. Adding a warning label in the interim is an easy band-aid, but it doesn’t address the underlying problem. Apple needs to do much more work here, and if it can’t, it needs to turn this feature off until it can release a version it can stand behind.

By Jason Snell

January 2, 2025 8:32 AM PT

Quick Tip: Which USB devices are currently attached?

I’ve recently switched from using two computers in two different offices (the hard-to-heat garage and my well-heated back bedroom) to using one computer (a MacBook Pro) docked in either location. There’s a lot that has gone into this decision, which I will detail in a future post, but this post is about one of the most frustrating aspects of this switch: the inability of my computer and automations and settings to understand when the context has changed.

Fortunately, my Mac reacts to switches to my network configuration with aplomb, so I don’t have to dive deeep into Network Locations, though if I did, there are several alternatives to old classic utilities like ControlPlane.

But I record a lot of podcasts, and I have two entirely different USB devices that I use for audio. How can I make it so that when I press the “record” button, the right USB device is recording and receiving audio?

My overall solution ended up being pretty complex, but the key insight was to use a Terminal command to list all the connected USB devices. You don’t need to know a lick of Terminal to use it, because you can stick it in a “Run Shell Script” block in Shortcuts. Here’s the command:

ioreg -p IOUSB -w0 | sed 's/[^o]*o //; s/@.*$//' | grep -v '^Root.*'

This command will output a list of all your currently attached USB devices in plain text.

(Optional explanation: ioreg will display an enormous list of devices and ports on your system. -p IOUSB restricts it to USB devices and -w0 makes it display complete devices one per line, without truncation. That result is sent to sed, which uses a regular expression to match just the names of the attached USB devices. The final step is to use grep to remove the initial line, which summarizes the list rather than listing an actual USB device. You don’t need to know this.)

At that point, you can build a Shortcut that alters its behavior based on the contents of the output. For example, my Shortcut’s next step is an If/then block that checks to see if the result contains the text “MV7” for my Shure MV7 microphone or “USBPre” for my USBPre2 audio interface. But it’ll work with any USB device that happens to be attached at a given time.

By Jason Snell

December 12, 2024 8:50 AM PT

Appearance: Apple Intelligence’s generic humans

On Wednesday iOS 18.2 and macOS 15.2 were released, and we wrote about it here. One of my complaints is that you can’t just make generic figures of people—you have to choose actual people in your library.

This is wrong. You can do it—I just completely missed the feature, because it wasn’t positioned or labeled in a way that made me understand what I was looking at. In the interest of correcting the record and also informing people whose brains work like mine, here’s the deal:

To create an image in Image Playground that doesn’t use the face of someone you know, click or tap the Choose… button with the word Person above it. This will bring up the person picker, full of faces from your Photos library. But in the top left corner of the picker is another option, Appearance.

The first time you click or tap Appearance, you’ll be prompted to choose a default appearance for your creations. (In subsequent creations, it’ll default to your previous choice, but you can change it by tapping Edit.)

In the Appearance view, you’ll be able to choose from five different skin tones and then from three different collections of Appearances. This is where it gets a little weird: Rather that building a person, Memoji style, you choose from three different collections of thumbnails, which are the various Appearances that might come up when you generate an image. They appear to be weighted by gender, so the leftmost is (mostly?) women, the rightmost men, the middle more of a grab bag—but there’s a lot of variation between each collection of Appearances.

That’s it. When you select a set, you’re ready to create an image. Unlike images generated from one specific person, you’ll find that different generations will be very different in this mode, because the entire appearance of the person can vary. As I swiped through a single set, I found young women, older women, young men, and people of various racial groups. It’s a grab bag. It’s meant to be generic. Go with it.

Once you’ve picked the Appearance, you can still add all the other prompts you want, either picked from Apple’s suggestion list or from your own terms. (I generated a bunch of starship captains with laser guns in space, enough to pack an entire star fleet.)

In Genmoji, instead of picking an Appearance, you pick an “Emoji” as a starting point. You can choose between the generic female, non-gender-specific, or male emoji templates, and choose a skin tone. All the genemojis you create will be based on that template.

By Jason Snell

December 11, 2024 10:57 AM PT

Last updated April 2, 2025

Mic Drop mutes your microphone everywhere

I’ve gotten so used to having a physical mute button on my podcast recording setup that it’s quite disconcerting when I use a different setup that doesn’t offer one.

In the winter months, I work most of the time in “Studio B”, a second desk in a back bedroom that’s climate controlled in a way my drafty garage can’t be. My microphone here is a sturdy Shure MV7, but I’m connecting it via USB and its mute button is a capacitive circle that’s super awkward to reach.

As a result, I’ve been trying to find a simple way to mute that microphone using a similar gesture—push to mute, push again to unmute—that I use in my primary recording setup. After exploring a bunch of options, I’ve settled on Mic Drop, a free ($5 upgrade for pro features) Mac utility that does the job perfectly.

Mic Drop literally does everything I expected from it. It lives in your menu bar and has support for global hotkeys or AppleScript, optional audio and multiple visual notifications of mute status, the ability to choose which mics are muted and which ones aren’t, and even an optional push-to-talk toggle mode.

It’s a delight to find an app that doesn’t just do the basics, but (via its recently-released 2.0 update) offers all sorts of polish that elevate into a utility that’s truly worth recommending.

MicDrop is available in the Mac App Store.

(Update: It now supports Stream Deck natively, too, which is awesome.)

By Jason Snell

December 6, 2024 5:12 PM PT

Simply brings its piano app to Vision Pro

I took piano lessons when I was a kid, and always hated practicing. I can blame it on the cold room we kept the piano in, but part of the reason I hated it was that most of what I played was boring. (I didn’t love having to take a long school bus ride to my piano lessons, either.)

This week I took Simply Piano for Vision Pro for a spin, and it was anything from boring. The popular iPad app for teaching piano has come to Vision Pro, and so I sat down at the very same piano I used to practice on as a kid—it’s in a somewhat warmer room now—but with a Vision Pro over my head.

Simply Piano works by listening to you playing notes and detecting if you’re playing the right or wrong ones. It’s very clever, but the Vision Pro version adds in the ability to overlay a virtual keyboard on your real one, so it can provide visual cues (in the form of glowing notes) when you’re not sure which key to play. It also annotates your fingers, so you can see which fingers are supposed to play which notes.

Even all these years later, I’ve got sight reading skills above Simply Piano’s introductory lessons, but as I went through the introductory lessons I found that the visual augmentation did feel a bit like magic. If I had any complaints, it was that sometimes Simply Piano struggled to recognize when I was playing notes, especially two of the same note in quick succession.

The Vision Pro app also comes with a virtual keyboard feature that allows you to practice on any flat surface, by overlaying a keyboard and letting you play it. I like this idea a lot, but in practice I found that it didn’t work very well. It played incorrect notes and even played notes when my fingers weren’t down. I think there’s something here—and having the ability to practice piano when you don’t have a keyboard handy is an amazing idea!—but I gave up pretty quickly and went back to my real piano.

Maybe if I had an iPad, or a Vision Pro, I would’ve practiced the piano more faithfully back in the day. Or maybe not. But this app seems to do a pretty great job of teaching the basics of piano, no teacher (or long school bus ride) required.