Meta’s Muse Takeover, Hallucinated Nukes, and Underwater Solar Panels | Tech News of the Week

[00:00:03.14] - Ned
Google seems to think for some reason that I care. And so at the top of my phone, it is now displaying the current score of the Eagles game.

[00:00:14.05] - Chris
Well, it's about time you joined the rest of society.

[00:00:20.03] - Ned
I don't know. I feel like there's probably a sport that's more popular than American football.

[00:00:26.06] - Chris
That's a ridiculous statement.

[00:00:29.23] - Ned
Do I immediately hear ICE knocking on my door making such American comments?

[00:00:37.10] - Chris
Maybe recording in the afternoon is also not a great idea.

[00:00:40.05] - Ned
Well, I think it's good for you. You have energy. But at this point, I've like lived a full day. I'm done. 1:45. It's fully time for me to Go to bed.

[00:00:54.08] - Chris
Yeah. Meanwhile, I'm just sitting here. I'm exhausted, just waiting for my groceries to be delivered.

[00:00:58.20] - Ned
Oh, you work so hard for it.

[00:01:01.19] - Chris
I had to choose between like 5 different kinds of Raisin Bran.

[00:01:06.05] - Ned
That's too many kinds of Raisin Bran.

[00:01:08.20] - Chris
I'm just saying my life is extremely difficult.

[00:01:12.07] - Ned
I can tell. Hello, alleged human, and welcome to the Chaos Lever Podcast. My name is Ned and I'm definitely not a robot. I'm a real human person with follicles, fallacies, and flux capacitors. We have those, right? Like at least 20 of them. With me is Chris, who's also here. Hi, Chris.

[00:01:32.00] - Chris
I think I remember, first of all, I kind of doubted the existence of Grape-Nuts for a long time. I was like, this is some kind of CIA psyop they're putting on TV to test the populace. Because why on earth would someone look at a product like that and say, I should have that in my pantry? Nay, my stomach.

[00:01:50.01] - Ned
I, you know, neither of those things are really the detractor about Grape-Nuts. It's the part where it's in your mouth and you have to chew it. Punish yourself. Grape-Nuts today. God, there's so many good— maybe we should work for the marketing department of whoever makes that. I'm sure it's General Mills or Post. I don't remember which one.

[00:02:10.18] - Chris
It's probably IBM. Who are we kidding?

[00:02:14.18] - Ned
Hey, speaking of IBM, let's talk about some tech garbage, shall we?

[00:02:21.02] - Chris
Everybody wants to kick Mark Zuckerberg.

[00:02:23.04] - Ned
Yo, yeah.

[00:02:24.06] - Chris
Wait, hold. I read that headline a bit too fast and skipped some important words. Let me try that again. Everybody wants to kick the tires on Mark Zuckerberg's new AI assistant Muse.

[00:02:37.12] - Ned
Eh. Hmm.

[00:02:39.06] - Chris
No, that first read still seems like the right read.

[00:02:42.19] - Ned
It does.

[00:02:45.07] - Chris
Anyway, fans of Muse, the popular band who I just learned was actually from England, woke up last week to find that their handles on Instagram, Twitter, and probably other places had been yoinked. @Muse no longer pointed to the guitar-heavy emo metal band it had for years and years. It now points to Muse, the Facebook AI assistant. The assistant went live Tuesday of last week, and while it hasn't been out for very long, It's definitely garnering attention. As of the time of the article in question, September 18th, 2026, Muse is the number one app on the Apple Store. Now, this isn't as impressive as it sounds. On day one, a download tracker site saw something like 81,000 downloads of Muse. Compare that to when Threads came out as its own app and hit something like 4 million downloads. But still, people seem to be using it. The App Store ratings are based on velocity, so something new and flashy often will hit number 1, at least for a short time. Muse, for example, is not nearly as high on the Play Store charts. So maybe there's some like New York Times top 10 bestseller bullshit going on.

[00:03:58.09] - Chris
I'm not sure. Anyway, Muse is intended to be a helper, not just a supercharged search engine. Which, let's be honest, is literally all LLMs will ever be. There are examples of people using Muse already to do things like unsubscribe from email lists, cancel unused subscriptions. One guy claims it saved him $3,500 a year on car insurance, which, dude, maybe don't pay for Muse, maybe pay for driving lessons. That's gotta be a lot of points is all I'm saying. Now, as the service goes, there is a free tier that has a limited number of tokens and slower priority of service. The token limit refreshes every week. Unlike actual Facebook, there is a pay version of Muse that claims it won't use your data for training, but since it's Facebook that's saying this, and since Facebook is currently being sued by Instagram users claiming Facebook used posts for training data and have been for ages, I'm not sure if I believe that claim. Time will tell if Muse is just a curiosity or if there is genuine utility and value here. I for one am going to sit here and be smug that an AI product started its life by stealing social media handles.

[00:05:15.01] - Ned
Dang, Muse the band must be pretty pissed they pulled all their handles. I mean—

[00:05:20.23] - Chris
I have not seen an official response from the band, which leads me to believe there has to have been some kind of an agreement. Of some kind, but I have a feeling it was more of a Darth Vader type of an agreement.

[00:05:33.02] - Ned
Yes. I, I, since Facebook or Meta holds the keys to everything, if they want your handle, they can really just take it. That's probably in the terms and conditions somewhere.

[00:05:43.10] - Chris
Yeah, I am sure that it is.

[00:05:45.19] - Ned
Nevertheless, Muse is not an unpopular band that people have never heard of.

[00:05:50.07] - Chris
Yeah. And apparently they do a lot with like Technology ruining society and dystopian type of music, which I think is very funny, all things considered.

[00:05:59.00] - Ned
How appropriate. Jev doesn't want to talk to you either. Hmm. A new player has entered the AI fray. This time it is TypeSafe AI emerging from stealth. Last week, TypeSafe AI announced a new model that is not a traditional LLM. They are calling it a System 1 model, borrowing from the taxonomy introduced in the book Thinking, Fast and Slow. Now, have I read this book? No. Why? Well, it feels very much like a Malcolm Gladwell book, and as all listeners of If Books Could Kill know, that's not a good sign. So instead, cribbing from the Wikipedia article, It appears that the book bifurcates our thought processes into 2 categories: System 1 and System 2. System 1 is thought that comes from the gut and largely feels instinctive, and System 2 is slower and more deliberate. The vast majority of decisions we make are System 1 decisions, because if they weren't, we'd never get out of the house in the morning, having been paralyzed by the frankly boggling number of choices for breakfast cereals in the cupboard. Seriously, why do I have 6 kinds of cereal? And what the hell are Grape-Nuts doing in there? Oh yeah, I had that health kick thing from last year.

[00:07:26.06] - Ned
I should probably go clean out my cupboard. What was I saying? Oh yeah, the new model. The new model from Typesafe AI is called Jev, and it's a system 1 decision maker, meaning it does not spend an inordinate amount of time thinking through a decision. Jev is deliberately not set up to be a chatbot and does not have a chat interface. Instead, you send it information and ask it a simple question, and it will respond in the correct format with a confidence value assigned. You can ask it to score something on a scale you define, answer simple yes or no questions, or ask it to classify something based on a set of labels you provide. So I could ask, Jev, should I throw out this 1-year-old box of Grape-Nuts I found in the back of the cupboard, yes or no? And I would get back an emphatic yes with 99% probability assigned. The model is designed to be lightning fast, and since it produces an output so rapidly, the output tokens from the model are free. This is in contrast to LLMs where output tokens are plentiful and they cost you money. Jev's average response time for queries is down in the 70 to 350 millisecond range, as opposed to several seconds or sometimes minutes for most frontier LLMs.

[00:08:48.06] - Ned
The announcement of Jev has made quite a splash on the AI scene, but I remain a little skeptical. How is Jev different from traditional sentiment scoring models of the past in the pre-LLM era? You know, this whole machine learning thing Actually existed for a while before we had ChatGPT. I don't know. It's like how Elon Musk tried to reinvent the train and call it the Hyperloop. We've seen how well that's gone.

[00:09:14.18] - Chris
Apple releases OS X, iOS, and iPadOS 27 and desperately hopes that version 26 never happened. On Monday, September 14th, Apple did something that, if I'm being honest, I was not expecting. They released an operating system that was better than the one that went before it. Now I know, I know. The 26 versions were rushed out the door because hitting a release date is now more important than actually having a product on the market that works. Thank you, capitalism. 26 was regarded as a disaster by most, with half-baked features and seemingly arbitrary changes to day-to-day operations. A lot of people, myself included, skipped 26 entirely. Those that bought new devices and had 26 foisted upon them complained loudly and constantly. It was a good time. Fast forward to, well, now, and 27 has come out, mostly to the relief of reviewers. It is being called boring. More stable and the least exciting update in years. And honestly, thank God. Someone referred to this release as, quote, finally fixing hundreds of paper cuts, unquote. And that is good because paper cuts suck. One challenge to note, a major release feature that's in heavy air quotes Is Siri AI?

[00:10:52.19] - Chris
That's right, 2 great tastes that taste great together. Siri, the widely mocked assistant that's been around for ages, is now becoming Siri AI, so it can still be incompetent, just more confident about it. We talked a little bit about this leaking into the Apple Watch release last week. Well, now it's everywhere. Yay. One thing that's interesting is that reviewers and beta testers are saying that the model, that powers Siri AI, Apple Intelligence, looks like it's been built in such a fashion that you can swap it out for something else, like an OpenAI or a Claude, which is interesting. Now, can you turn off Apple Intelligence and Siri AI in 27, you may ask? And the answer is yes, you can, but it's a pain. First, you have to change Siri to Siri Classic and then disable summarize features in every app that has Apple Intelligence shoehorned into it. How fun. And surprisingly, disable Apple Intelligence makes up like the first 15 search results when you search for Siri AI. Super popular.

[00:12:02.01] - Ned
Totally.

[00:12:03.04] - Chris
One other thing, to even turn on Siri AI is going to require, eh, not the cutting edge, but modern hardware. To get the advanced features to work, you will need M3 chips or better on desktop. And iPhone 15 Pro or better to get any features at all in your pocket. 27 has no support at all for Intel CPUs, so pour one out for those. Oh, and at least for now, there's still a waitlist. So even if you click to enable, it's not a guarantee you'll have Siri AI features right away. I am guessing it's not going to be that long of a wait. Now, to be fair, there are positive reviews of Siri AI out there. The AI Chronicle, which I'm sure is an unbiased source, refers to Siri AI as, quote, the assistant that finally understands. Siri AI has the capability to observe absolutely everything that's happening on your screen as well as in the apps you're running. Now, for me personally, this is a deal breaker because gross. All of the number crunching goes on in Apple's private cloud compute, which we've name-checked enough times that Ned should probably just write an article about it.

[00:13:15.06] - Chris
So anyway, there you have it. OS X 27, the best OS release from Cupertino in years, and its biggest feature is that it isn't 26. So long as you're going into it eyes open about the Siri AI stuff, this is probably a worthwhile upgrade.

[00:13:34.03] - Ned
Why do I have a feeling we'll be doing the same story when Windows 12 finally comes out? Hallucinating AI almost destroys humanity. You ever get the feeling that the apocalypse will not be triggered by some shady cabal or an intricate subterfuge, but rather just a stupid mistake or simple misunderstanding? Considering how close we've come to thermonuclear war in the past due to a moonrise over Norway, a failed relay station in Colorado, Or a particularly strong solar flare, all real instances, all very dumb. Well, I guess you could be, you could understand why we would think this might happen. And in that vein, the US came close to forcibly boarding a Chinese ship in search of nuclear material that was reported to be present based on an intelligence report. Spoiler, it was not there.

[00:14:30.12] - Chris
Probably.

[00:14:31.00] - Ned
Fortunately, we didn't board, which would have caused an international incident and maybe sparked, you know, World War III. The report in question was generated by a US security analyst with assistance from AI. The analyst in question had used an unnamed chatbot to summarize the contents of the Chinese ship's manifest. It would appear that the model hallucinated the presence of banned nuclear arms on the ship. Which probably should have warranted a bit more scrutiny from that analyst. Instead, the report was distributed, leading the Navy to prepare for active military aggression against the Chinese. Fortunately, the source of the document was discovered before the action was executed. As of late, the U.S. military has started to heavily use AI for multiple activities. That includes models from Google Gemini, Anthropic, and most recently Grok. Now I'm not saying this was probably Grok's fault, but I'm also not not saying that. Actually, Gemini is pretty terrible as well. Still, saying that the AI fucked up your homework is not a valid excuse for a student or a military officer, and I hope the analyst was chastised for their shoddy work. As someone who uses AI on a regular basis to do research and summarization, I treat the results much like I treat stories my buddy Matt used to tell me.

[00:15:57.22] - Ned
There are— they are certainly grounded in reality but are prone to flights of fancy. Like, I believe that Matt once worked in a pool hall and used to get tips under the table. I'm slightly more skeptical that he once beat Taylor Swift in a game of 9-ball. As we all know, Swift is a pool shark in addition to being a world-class entertainer.

[00:16:18.09] - Chris
Scientists build solar panels that can work Underwater. So here's a fun fact that you may or may not know. Earth, you see, is mainly covered with water. There's a lot of it.

[00:16:31.22] - Ned
Whoa.

[00:16:32.00] - Chris
In fact, we should probably call it Water, not Earth. Wow, that joke's awful. Exploring the ocean is not unlike exploring space in that reliably generating power for long periods of time without carrying excessive weight is a big challenge. Recently, researchers in China have started to solve this problem in a surprising way, via solar panels. Now, solar panels on the surface have made amazing progress over the past 2-odd decades, generating lots and lots of power at a reasonable cost. It's been nuts. It's something like a 100x improvement in terms of the amount of money it costs over the past 20 years to generate a watt of electricity.

[00:17:15.17] - Ned
Mm-hmm.

[00:17:16.07] - Chris
But traditional solar panels are wildly inefficient and susceptible to damage in ocean-type conditions. So researchers have been looking at a different kind of solar cell based on a compound called perovskite. Researchers were able to tweak the mixture of perovskite into a material that can absorb the dim and differing wavelengths of light that reach underwater as well as resisting moisture and salt-based damage that would doom any traditional panel. Scientists tested this both in a lab environment as well as out on the real ocean and reported that about 35% light into electricity ratios were possible at 10 meters below the surface.

[00:17:59.17] - Ned
Wow.

[00:17:59.21] - Chris
This is way better than silicon-based panels that would be more like in the 20% range. Moreover, the material science that went into this Means that these panels could last for more than 5 years. A working solar panel array that floats below the surface at 10-odd meters could be used to power submersibles that are tethered to it and going far, far deeper and staying far, far longer than if they were powered by batteries. I was looking for a zinger to end this one, but I actually think this is just a cool story. I am famously not a water person per se. I have the swimming aerodynamics of a dolphin made of concrete, but I still like the idea of science having the ability to do long-term research deep underwater where we've never— this is going to be war, isn't it? This is going to end up being a war thing.

[00:18:57.06] - Ned
Internet Explorer will finally die in 2029. Sort of. It's hard to believe, but there was a time when Internet Explorer ruled the browser landscape. It was a dark time filled with ActiveX exploits, unsupported web standards, and Flash plugins eating all of your system RAM. Also, Silverlight was a thing, almost. Peaking in 2004 with almost 96% of the browser market share, Internet Explorer was swiftly toppled by Firefox and Chrome, with Chrome becoming the winner sometime in 2012. Even though Chrome has been the dominant browser for over a decade, there were a lot of web applications written in the early aughts and 2010s that leveraged IE's unique features. I'm not going to call them bugs. These web applications live inside the corporate firewall, Where there is little appetite or budget to rewrite those apps to meet the current web standards of today. With support for IE ending in 2022, what was Sharon's Swiss Cheese Supply in Sheboygan, Wisconsin to do about their internal app, Cheese Counter XP? Dave, the lead developer, retired in 2016 to a dairy ranch and has sworn off all technology, as we all should. Living the dream, Dave. Enter IE mode for Microsoft Edge.

[00:20:26.06] - Ned
When running in IE mode, the Edge browser emulates some of the features and functionality of the Internet Explorer browser, lying to the web application about which browser is making the requests. While it doesn't work for all ancient IE web apps, it's an avenue to extend your use for a few more years. Unfortunately, that dream too has come to an end, and it might be time to coax Dave out of retirement for one last hurrah.

[00:20:52.14] - Chris
Don't do it, Dave.

[00:20:54.08] - Ned
According to a Message Center advisory from Microsoft, IE mode for Edge will be supported through 2029. Now, of course, Microsoft has been known to extend support deadlines provided a company is willing to pay enough money, but this may finally be the wake-up call to anyone running an ancient web app on IIS 6 on a server sitting under Dave's old desk. The time has come to embrace modern web standards. Or you could just run a Windows XP emulator and live without support. It's your call, Sharon. Your call. Hey, thanks for listening or something. I guess you found it worthwhile enough if you made it all the way to the end. So congratulations to you, friend. You accomplished something today. Now you can go fire up Internet Explorer 7 and watch videos on Newgrounds for the rest of the day. You've earned it. You can find more about the show by visiting our website, chaoslover.com, where you'll find show notes, blog posts, and other tomfoolery. We'll be back next week to see what fresh hell is upon us. Ta-ta for now. Like, if you've ever enjoyed going to the beach and getting sand in your mouth and chewing through it, Then you are a great candidate for Grape-Nuts.

[00:22:16.16] - Chris
Wow. I mean, that, that's a marketing, that's a marketing coup right there. That's just, that's just brilliant.

[00:22:23.05] - Ned
It's like chewing kitty litter, but better.

[00:22:25.15] - Chris
Oh, that's— you were, you were riding so high for a second there.