Archive for September 18, 2023

Monday, September 18, 2023

iOS 17

Apple (release notes, Hacker News):

iOS 17 makes iPhone even more personal and intuitive with major updates to communication apps; StandBy, a new way to experience iPhone when it is charging; easier sharing with AirDrop; and more intelligent input that improves the speed and accuracy of typing.

See also: Federico Viticci, Dan Moren, MacRumors.

Ken Case:

Late in the iOS/iPadOS beta cycle, a memory issue was introduced in Apple’s JavaScriptCore framework which can (but doesn’t always) trigger crashes when JavaScript code is used in an app. Since Omni Automation plug-ins and scripts are all written in JavaScript, this can cause our apps to crash any time they load or run code from those plug-ins and scripts. This doesn’t happen all the time, but it happens enough to be bothersome.

We’ve consulted with Apple about this intermittent JavaScript crash, and hope it gets resolved soon.


Apple’s New Transformer-Powered Predictive Text Model

Jack Cook (via Hacker News):

The feature will occasionally suggest more than one word at a time, but this is generally limited to instances where the upcoming words are extremely obvious, similar to the autocomplete in Gmail.


I have to say that this vocabulary file strikes me as pretty unique, but it’s definitely not out of the question for a language model deployed in this setting. I’ve personally never seen emojis featured so prominently in a language model’s tokenizer, but existing research has shown that domain-specific models and tokenizers can drastically improve downstream model performance. So it makes sense that a model trained for use in things like text messages, in which emojis and contractions will be used a lot, would prioritize them.


GPT-2 has four main parts: token embeddings, positional encodings, a series of 12-48 decoder blocks, and an output layer. The network described by unilm_joint_cpu appears to be the same, except with only 6 decoder blocks. Most of the layers within each decoder block have names like gpt2_transformer_layer_3d, which would also seem to suggest it’s based on a GPT-2 architecture.

From my calculations based on sizes of each layer, Apple’s predictive text model appears to have about 34 million parameters, and it has a hidden size of 512 units. This makes it much smaller than even the smallest version of GPT-2.

The early reports about auto-correct in iOS 17 and macOS 14 seem to be positive. I’m cautiously optimistic that it will fix the biggest problems for me with the old system, which are that it suggests words that are not spelled correctly and even changes correct words that I entered into mistakes.


Maps in 2023

Justin O’Beirne (Hacker News):

For the past three years, Google’s cartography has largely remained in this difficult-to-scan state—that is, until now. That’s because as of late August 2023, Google appears to be testing a new Apple Maps-inspired map style.

Unfortunately, Google’s new, in-testing map style is even worse than its old one. Here in Chicago, for instance, notice how much harder it is to read and scan the map.

I have always found Google Maps easier to read than Apple Maps. I prefer the coloring and how it chooses which roads and details to show when. Apple Maps overemphasizes showing businesses, and it typically picks a few irrelevant ones to highlight while hiding all the others. The differences are most pronounced in CarPlay navigation, where Apple Maps shows large signpost-style street labels rather than drawing them on—and oriented to—the actual roads. Apple Maps also covers large portions of the map with status information. Unfortunately, Google’s map style has been gradually moving in the direction of Apple’s design, and these changes continue that trend.

In many ways, I think Apple Maps works better as an app, though. And I like how it can show an imminent turn on my car’s dash board and on my Apple Watch. Despite this, I prefer Google Maps because of the way it displays the maps and because overall Google still seems to have more accurate place information. That said, I am now in the habit of checking both maps sometimes because I recently had a disastrous experience where Google Maps routed me on a road that didn’t exist, which I didn’t find out until I was there and out of cellular range. When I got home, I checked what Google’s second-choice route would have been, and it recommended another road that didn’t exist. Apple Maps had the correct information in this case.

Michael Grothaus (via Hacker News):

Google Maps still holds around 80% of the mobile market. But in recent years, I’ve found myself getting increasingly frustrated with the Google Maps experience, especially when it comes to general navigation and exploration of a map area.

Here are the five main reasons Google Maps has become a cluttered, frustrating mess—and why I find myself turning to Apple Maps more often.

Joe Rosensteel:

On the most recent episode of Accidental Tech Podcast, Marco Arment described a recent ordeal with a moving truck. He recounted his list of grievances from using a rented box truck in New York, and mentioned the issue of roads that trucks aren’t allowed on, or won’t fit on. New York parkways (park on the drive way, yeah, yeah we all know the joke) apparently don’t allow trucks. The major mapping applications he mentions - Google Maps, Apple Maps, Waze - don’t have a feature to avoid roadways that don’t allow trucks, as they do for avoiding tolls.


If you were in that same 105 E HOV 2+ lane, with two passengers, and no transponder, heading from LAX to Downtown LA you would need to exit the HOV lane and merge on to the 110 N at the right. If you stayed in that HOV 2+ left lane because you didn’t understand the signage, you would be dumped into the FasTrak lane that requires a transponder.

I’ll leave it at just those examples, because you get the idea. These paricular lanes have been like this for over a decade. Other lanes like this exist elsewhere, and more are being completed right this very minute.

Apple and Google are totally clueless about these lanes, which is bizarre when you can see them represented in maps, satellite views, Street View — everything.


Toyota Factory Shutdowns

Justin McCurry (via Hacker News):

The Japanese carmaker said the stoppage on 29 August at all 14 of its domestic plants occurred after servers that process orders for vehicle parts broke down following a maintenance procedure carried out the previous day.

During this operation, “data that had accumulated in the database was deleted and organised, and an error occurred due to insufficient disk space, causing the system to stop”, Toyota said on Wednesday.


Toyota had to shut down the same 14 factories for a day in February last year when one of its suppliers said one of its file servers had been infected with a virus, raising questions about the cybersecurity of Japan’s supply chains.

Andrew E. Freedman:

Moral of the story: Always keep some extra storage space on hand for anything mission-critical, especially if you’re a massive company that requires software that can be timed to exact specifications in order to build complicated machines.