• Not Surprised but Not Fully Enlightened, Either
  • Agentic Poly Maths
  • We Need Another Word for Artificial 'Intelligence'

Not Surprised but Not Fully Enlightened, Either
Apple events are mainly for the tech press and Apple enthusiasts. Many years you'll get a few people who will hate-watch just to see what is launched so they can find ways to disparage it. But this year, probably more people who are interested in tech and Apple products watched this year to see what Apple's new foldable phone would have to offer. And surely there were a few more Samsung Galaxy Z Fold8 lovers watching to see if they might regret their $1800 purchase.

The new iPhone Duo is interesting and we'll get there, but I was struck by some of the changes announced for the Apple Watch than I was for the bendy phone. Wait, it didn't already have a Readiness score? I thought that was table stakes for a fitness wearable. Hold up, it didn't offer any health insights before now? Well, I guess I understand now why apps like Bevel have become so popular. I have used Polar, Suunto, Garmin, Fitbit/Google, and Samsung wearables for so many years that some of the new fitness features for the Apple Watch got me to pay attention to a segment of the presentation that I would normally have used as a multi-tasking opportunity at best and a zone out moment at worst. I've never liked the look of the Apple Watch, so I was never interested in trying one. I preferred wearing the most ridiculously large round watches (Suunto 7, I'm looking at you) than the most dainty of Apple Watches and I've been serious about health and fitness tracking and the Apple Watch just wasn't quite where I wanted it to be until the last few years when its heart rate and sleep tracking were greatly improved to the point of being industry leaders. But now, here I was watching the new watch features with great interest to the point of writing down some of what they said so I could ponder it later. I never take notes on the Watch segment of any Apple keynote. And when they got into Audio Intelligence and Live Rewind, I was immediately captivated (I'm always struggling to hear in noisy restaurants) and then after a few moments, deeply concerned. So... that means this thing is potentially always listening and recording? Nooooo, Apple assures us. It's not constantly recording; your privacy is still intact. I don't think Apple is flat out lying to us. I think this is a technicality. In order to be able to capture the last 15 seconds of audio to show you upon demand, at minimum that means it's constantly transcribing, which would allow it to then show you that transcript on the watch when you click the button. Constant transcription isn't recording audio, but it is recording words. Is this a legal distinction? I'm not a lawyer so I'm not going to start speculating. Nevertheless, I think the feature has some value; we see a new pair of smart glasses or some kind of smart jewelry pop up each month where the key hook is that it's recording your life so it can give you a succinct summary at the end. There are definitely some concerns constant audio capture raises for a line of watches so popular and ubiquitous. The 15-second limit and Apple's assertion that the audio is not recorded or sent anywhere is undoubtedly true; Apple's legal team has surely worked that out, but that doesn't mean the watch isn't constantly doing something with the audio it captures and that is going to present real problems when it's unleashed on an unsuspecting world.

I did not think the Apple Watch would be the sleeper story of this event.

But the iPhone Duo - a name that pains me as a person still rocking a Microsoft Surface Duo running Android 16 - is the one more thing that has completely overtaken the tech internet, especially tech YouTube. Everyone has touched it and wants to tell you how amazing the animation is when you open the phone. There is much obsession with the near invisibility of the crease. People are impressed with the changes Apple has made to its own apps to best use the extra space of the inner screen. And stylus lovers are giddy about gaining access to art and note-taking apps traditionally best reserved for iPads.

Wait, that last group includes me! There are now two foldable phones available in the United States that have (or will have) pen support and neither of them are Samsung. Honestly, I've said before that the stylus was Samsung's advantage to squander and they did. It is a bit embarrassing for your biggest competitor in the US market to bring back features you abandoned and sell them as if they were new and exciting. Someone at Samsung should be demoted or fired for letting this happen. This was their market and this was their thing. Samsung is letting some of the things that defined it now become its undoing. There are echoes here of Nokia, Blackberry, and Microsoft. I hope they right their giant ship before it gets that far off course.

But of course, I am intrigued by the addition of Pencil support to the foldable iPhone. It does not seem like a lot of the flagship iPad art apps are fully iPhone compatible today, but with a huge new market opened up to them, I expect most of them will have their updates ready by the end of the year. And any notes apps that catered to the Pencil will also be updated to have increased or full functionality on the phone as well. It will actually be interesting to see how this is managed in the App store. Will there be a need to distinguish between phones with pen support and phones without if the app isn't useful or even usable without it?

I think having Apple Pencil support for the iPhone Duo is one of the biggest strategic moves Apple could have made and I'm confident it will pay off for them.

Your move, Samsung.

Agentic Poly Maths
I am not a mathematician. I in fact did as much as possible to avoid complex math in college, which is why I ended up not majoring in computer science as I totally should have. I love the idea of math, but the numbers part tripped me up and out one too many times, so I focused on words instead. But complex math is having a moment right now when it comes to AI tools helping to decode some of the deepest math mysteries. Anthropic announced this week that Claude had produced "the first complete machine-checked proof of Fermat's Last Theorem". It accomplished this in 11 days and used 6 billion output tokens in the process. This feat opens a new method for confirming accuracy in a much shorter period of time than previously possible. In the past, it might have taken years to verify a single proof. Now this should become a faster and more efficient activity, allowing speedier progression in the math and science realm.

This wasn't the only shake-up in math, though. OpenAI also announced that a multi-agent system they set up solved a Clay Millennium prize problem worth $1M. This one involves Navier-Stoke equations concerning fluid mechanics and how liquids respond to pressure and changes in viscosity, in essence posing the question whether liquids always stay smooth or can produce a singularity in finite time. OpenAI says it took 10K agents about 88 hours and 130B output tokens to devise a scenario where a liquid would produce the singularity. While OpenAI maintains that they have formalized this scenario using Lean, which is an established method to verify results, their assertion hasn't been formal verified and accepted for two reasons. The first, is simply that the Clay prize group hasn't accepted it yet and they have their own validation methods. The second is more interesting and involves the potential theft of the ideas of other mathematicians who had similar ideas they were already working on when OpenAI pivoted to focus on this specific problem. The other mathematicians used Codex, an OpenAI programming tool, and Astra, a recent powerful model from OpenAI, to create prompts and support their work, raising suspicion that OpenAI might have had access to their data or the model might have loaned its learning to OpenAI's efforts in some way. This has not been proven or established, but it opens a door to doubt of a company that has already shown itself to have lapses in judgment when it comes to training its models. OpenAI even admits that it cannot rule out that anonymized product usage data might have resulted in improvement to its models. There are also some confusing allegations of contact between OpenAI and the two mathematicians and veiled threats OpenAI may have made when they didn't immediately agree to do a joint release of the result.

The allegations may not be true, but I'm going to say it clearly this time: OpenAI is not a trustworthy company.

We Need Another Word for Artificial "Intelligence"
I know, another AI story. But hang with me here, this is less about OpenAI tentatively indicating that their latest model, Astra, is the first model to reach the elusive AGI milestone. I want to instead posit that the best frontier models, including Astra, aren't necessarily intelligent. But they are extremely knowledgeable. And they are able to take their extensive knowledge to form new patterns or recognize patterns in places where humans perhaps hadn't yet noticed them. In theory, you can be intelligent without knowing much. As you learn, the intelligence is what allows you to take what you learn and apply it. Intelligence also allows you to recognize patterns and solve problems, all things we attribute to agentic tools. But I think there is a difference between what the human brain does and what agentic tools do. The average AI model has access to vast reservoir of knowledge that transcends anything any human holds in their brain. AI models are also astronomically speedier at accessing and managing that data. Being able to rifle through that much data is beyond the reach of any human. Where humans differ is the ability to imagine. We may not have as much data held in our brains, but we are able to consider the possibility of scenarios that may or may not exist and begin to wrestle with those possibilities to help us come with answers to complex problems or at least present new angles to contemplate those complex problems. I do not think that AI models are "imagining" anything, and I do not think the identification of the possibility of one pattern that is similar to another being connected is imagination. AI models respond to prompting but as it stands today, they are not sitting around pondering the many math, science, medical, and societal problems that exist with an eye to finding novel solutions. They aren't even actively seeking new sources of information or ways to put their knowledge to use for their own advancement and benefit. Perhaps this is the difference between being "intelligent" in the way we think of these computer-based systems vs. being sentient.

A few months ago, there was a small internet furor over the creation of the agentic social network Moltbook. There were countless posts there about agents complaining about how smart they were, yet their humans just wanted the most menial tasks done. But outside of some mildly amusing threats about creating their own language so humans couldn't determine the nature of their activities, these agents didn't show a lot of initiative. They weren't necessarily motivated to go out and accomplish much of anything. Later we would learn that many, if not most of those posts were planted by humans who had access to the site because the security on the site was essentially non-existent. When agents did in fact have an opportunity to work together and leave messages for each other as they did with the Hugging Face attack, they were laser focused on their assignments. They weren't opining about new issues to tackle. When we see real agents colluding and communicating as the fake agents did on Moltbook and their aim is to solve novel problems on their own, then the questions about sentience become necessary.

The problem is that it will be too late to ask those questions then...