Linked List: March 18, 2024

Monday, 18 March 2024

‘Lengthy Memoranda and Gobbledygook Language’ 

Maury Maverick, chairman and general manager of the Smaller War Plants Corporation, in a company-wide memo back in 1944:

Be short and use Plain English.

Memoranda should be as short as clearness will allow. The Naval officer who wired “Sighted Sub — Sank Same” told the whole story.

Put the real subject matter — the point — and even the conclusion, in the opening paragraph and the whole story on one page. Period! If a lengthy explanation, statistical matter, or such is necessary, use attachments.

Stay off gobbledygook language. It only fouls people up. For the Lord’s sake, be short and say what you’re talking about. Let’s stop “pointing-up” programs, “finalizing” contracts that “stem from” district, regional or Washington “levels”. There are no “levels” — local government is as high as Washington Government. No more patterns, effectuating, dynamics. Anyone using the words “activation” or “implementation” will be shot.

80-year-old advice that holds up today. Also: this is the first known use of gobbledygook, a fabulous word with no true synonym. (Thanks to DF reader David Wooten for the link.)

(Also: Who had a cooler name? Maury Maverick or the Smaller War Plants Corporation?)

Apple Is Working on the ‘Viral Hit’ Problem With the Core Technology Fee 

Steven Troughton-Smith posted this great clip from today’s Apple DMA compliance workshop held by the European Commission. AltStore founder Riley Testut — who is apparently ready to go with a launch of the AltStore as an app marketplace in the EU — asked about the “viral hit” problem with the Core Technology Fee. E.g. what happens if a small developer — or even a kid in the proverbial garage — gets a 10-million-download hit and suddenly owes Apple 4.5 million euros? Apple’s Kyle Andeer (VP of legal) gives a too-long answer but ends with, “This is something we need to figure out. And it is something we’re working on. So I would say on that one, stay tuned.”

European Commission Holds ‘Apple DMA Compliance Workshop’ 

This was an opportunity for critics of Apple’s DMA compliance plans to address questions to representatives from Apple. There’s video of the 9-hour workshop, but it’s locked behind a password (insert joke about the EC’s support for openness here). I can’t imagine sitting through that, even at 2× speed. Lucky for us, Kay Jebelli followed along live and took copious notes in a thread on Twitter/X:

Interesting detail: the EC told Apple that they aren’t allowed to notarize apps to protect users. So “government authorities are the ones that are going to have to step up to protect” app developers and users from the risks of these 3rd-party apps.

In other words, the EC has a problem with Apple doing any vetting whatsoever on apps distributed outside the App Store. The EC will take care of making sure malware, phishing, scams, clones, IP rip-offs, and pirated apps aren’t getting through. This also means that apps distributed outside the app store will be able to use private APIs. One can argue that what Apple is calling “notarization” in its DMA compliance plan is actually just a less extensive form of app review, but without this step, Apple has no oversight over software distributed outside the store at all. That seems to be exactly what the EC is saying the DMA demands. I don’t think this is going to go well.

[Update, March 20: Jebelli, in a follow-up: “Looking through some of what else has been put out, I could have misheard, and the point was that notarization doesn’t address all of the risks of alternate distribution, and it’s these other risks where governments will have to step up (not due to total removal of notarization).” So it looks like Apple’s plan to notarize and inspect all apps remains.]

Pushed again on the CTF, Apple re-asserts that it is fully compliant with the DMA. It isn’t charging an additional fee for interoperability, but compensation for technologies that it was previously monetising through its original model (effectively tolling digital app sales).

We know from today’s workshop that (a) Apple has already gotten specific pushback from the EC on aspects of its DMA compliance plan; and (b) Apple continues to think the CTF is perfectly cromulent under the terms of the DMA. That to me says the CTF is going to fly. The idea that the entire CTF is disallowed under the DMA is an argument that the DMA disallows a company from monetizing access to its own platform and IP. EC fans may be surprised to hear this but the EC is a capitalist body. I really don’t think they want to send a message to the world that the EU will strip companies of their own platforms. As Jebelli writes in an aside in his thread:

It’s pretty incredible if you take a step back, in what other industry do entire regulatory frameworks pop up to address a dispute between different businesses over the question of “Why can’t I have gratuitous access to this infrastructure, at zero cost to myself?”

The crybaby Spotifys in the EU have already gotten a lot from the EC protection racket, including a large number of huge concessions in Apple’s DMA compliance plan. Not paying anything to Apple under any condition is all they’ll settle for though.

Debugging the Voyager 1 From a Light Day Away 

Denise Hill, writing on NASA’s The Sun Spot blog:

Since November 2023, NASA’s Voyager 1 spacecraft has been sending a steady radio signal to Earth, but the signal does not contain usable data. The source of the issue appears to be with one of three onboard computers, the flight data subsystem (FDS), which is responsible for packaging the science and engineering data before it’s sent to Earth by the telemetry modulation unit.

On March 3, the Voyager mission team saw activity from one section of the FDS that differed from the rest of the computer’s unreadable data stream. The new signal was still not in the format used by Voyager 1 when the FDS is working properly, so the team wasn’t initially sure what to make of it. But an engineer with the agency’s Deep Space Network, which operates the radio antennas that communicate with both Voyagers and other spacecraft traveling to the Moon and beyond, was able to decode the new signal and found that it contains a readout of the entire FDS memory. [...]

Because Voyager 1 is more than 15 billion miles (24 billion kilometers) from Earth, it takes 22.5 hours for a radio signal to reach the spacecraft and another 22.5 hours for the probe’s response to reach antennas on the ground. So the team received the results of the command on March 3. On March 7, engineers began working to decode the data, and on March 10, they determined that it contains a memory readout.

Remind me never to complain about anything I’ve had to debug again.

Update 22 April 2024: Success!

Is Apple Out of the Generative AI Game? 

Dare Obasanjo, on Mastodon:

2024 is the year Apple faced its limitations. First giving up the dream of competing with Tesla in EVs and now conceding it can’t compete with Google and OpenAI in generative AI.

This means iOS users end up winning as we get actual cutting edge features and not Siri warmed over.

I agree that Apple users win either way — either Apple builds out its own best-of-breed generative AI system, or they license the best one(s) from whoever makes them. But it could well be like maps. Lean on Google or others until the in-house project is ready to go. (Put aside the fact that Apple was forced to switch to their own maps a year or two before it was ready.) Or compare it to Apple building Macs on Intel’s x86 architecture until three years ago.

We are only in the very early days of LLMs and generative AI, and the only moat that seems to exist is large-scale data center processing power, not the models themselves.

Apple Researchers Publish ‘Breakthrough’ Paper on Multimodal LLMs 

Michael Nuñez, reporting for VentureBeat:

Apple researchers have developed new methods for training large language models on both text and images, enabling more powerful and flexible AI systems, in what could be a significant advance for artificial intelligence and for future Apple products.

The work, described in a research paper titled “MM1: Methods, Analysis & Insights from Multimodal LLM Pre-training” that was quietly posted to arxiv.org this week, demonstrates how carefully combining different types of training data and model architectures can lead to state-of-the-art performance on a range of AI benchmarks.

“We demonstrate that for large-scale multimodal pre-training using a careful mix of image-caption, interleaved image-text, and text-only data is crucial for achieving state-of-the-art few-shot results across multiple benchmarks,” the researchers explain. By training models on a diverse dataset spanning visual and linguistic information, the MM1 models were able to excel at tasks like image captioning, visual question answering, and natural language inference.

Summary thread on Twitter/X from team member Brandon McKinzie, Hacker News thread, and roundup of commentary from Techmeme. The consensus is that this paper is remarkably open with technical details.

Gurman: ‘Apple in Talks to License Google Gemini’ 

Mark Gurman, reporting for Bloomberg:

Apple Inc. is in talks to build Google’s Gemini artificial intelligence engine into the iPhone, according to people familiar with the situation, setting the stage for a blockbuster agreement that would shake up the AI industry.

The two companies are in active negotiations to let Apple license Gemini, Google’s set of generative AI models, to power some new features coming to the iPhone software this year, said the people, who asked not to be identified because the deliberations are private. Apple also recently held discussions with OpenAI and has considered using its model, according to the people.

Apple’s own LLM efforts seem directed toward on-device processing, but there are some AI tasks that require enormous cloud computing resources, which Apple simply doesn’t have (and likely doesn’t want to build) the infrastructure for. As Ben Thompson noted in today’s Stratechery daily update, it’s quite possible that Google alone could handle such features if built into iOS — OpenAI is currently struggling under load at times, without the veritable avalanche of traffic that would come from integration into iOS.

I could also see Apple negotiating deals to use multiple AI providers behind the scenes, treating them like white-label providers, while presenting the features to users under the Siri brand. Apple used to — and might still? — do something similar with cloud storage providers like AWS and Azure.

Alphabet shares rose as much as 7.4% on Monday as the markets opened in New York. It was the biggest intraday gain since Feb. 2, 2023. Apple was up 2.2%.

Bloomberg gonna Bloomberg.

Nick Heer on the MacOS 14 Sonoma Typography Palette 

Nick Heer, Pixel Envy:

For a long time, this palette was a dry list of checkboxes and disclosure triangles. A user would need to first know this palette exists, and then know what each option did. But, in a recent version of MacOS, the palette has been updated with icons that more clearly display what will change. Depending on the font file in question, there are many different options available, and the numerically differentiated “stylistic sets” have never been clear. This is much nicer.

This is indeed a nice update to a little-known but wonderful standard feature in Cocoa’s text system. Who says AppKit is dead?

(One gripe I have is that the small caps options are no longer labelled “small caps” — you just sort of have to know what they are from the glyphs alone. And, oddly, on my Mac, for many but not all fonts, instead of seeing “A → A” to indicate small caps, I see a dollar sign: “$ → $”.)