Don’t judge AI on what it could be—judge it on what it is: The Week in Review
Revenue is utility.
In case you missed it, check out our most recent long-form essay, “In praise of the shutter button.”
The Verge has long had a policy to review the product that ships, not the future potential of said product. It’s a good policy, especially when it comes to tech, as the potential of a product is often far more compelling than the thing you’re holding in your hand.
The $199 Rabbit R1 is a perfect example. I am a sucker for Teenage Engineering’s industrial design, and so I ordered the R1 simply because it looked great. I thought, how bad can it be?
The answer is astonishingly bad. It’s worse than useless; it’s not fit for purpose. The R1 does nothing that my iPhone or Apple Watch can’t do infinitely faster and more accurately. I should have known better, but at least I didn’t make the same mistake with the Humane pin, which is all of the above but costs $699 plus $24 per month subscription. Humane are, as the English say, are “havin’ a laugh,” which roughly translates to “they’re fucking with us.”
So the $600 billion question is: When are we going to start applying the same rule of judging AI products with the same “review what’s in the box” approach we apply to reviewing AI hardware products?
The more we read and write about AI in general, the more the common throughline seems to be OpenAI et al stealing data from creatives to train the LLMs with the promise that AI will be the most useful game-changing technology ever invented by humanity. OK, but when? Because if you want me to pay $20 per month for Gemini Advanced or ChatGPT advanced, what exactly am I paying for and why do I need it?
I have already canceled the renewal of my subscription to Midjourney and, to date, I have used only 20 of the 5000 credits I bought to use on OpenArt. Why? Apart from the ethical considerations that Ross and I have written about ad nauseam, all of the results I have created using either platform just look like derivative AI; I would have to do so much work to make the AI results useable or useful I might as well do it myself from the get-go.
I know some creatives using AI to remarkable effect—we are talking to a few of them about their processes, more about that in the coming weeks. But personally, I am tired of reading that AI will get better and become more useful. Will it, though?
I know it’s entirely logical to assume—considering the eye-watering amount of money AI companies are raising and the incredibly smart people working on developing these mind-bending complex tools—that there is a huge future business here. But as we all know, when it comes to humans, logic rarely has anything to do with anything. I question the assumption that millions (billions?) of people will pay for using AI. Yes, you like using ChatGPT, but would you cancel your Netflix subscription to afford it? Hell no!
I freely acknowledge that many people find ChatGPT, Gemini, Llama, Perplexity, etc., useful tools. I am also aware that there is massive potential for AI in the B2B space, and the impending launch of Apple Intelligence will bring AI to billions of iPhone users—emphasis on “impending.” See below.
But, useful and potential aren’t close to being enough to justify the ongoing amount of investment required and the ethical price currently being paid. To put a finer point on it, billions upon billions of investment will be needed to sustain something that, at present, is nothing more than a novelty. That tipping point will only occur when AI ceases to be just useful and becomes indispensable for billions of people.
Yes, that may happen one day. Yes, OpenAI and other AI companies may become as indispensable and powerful as Apple, Microsoft, Google, Amazon, Adobe, etc, are today. It is certainly possible, maybe even probable. But as it stands right now, after reviewing the product that is currently shipping, we’re not even close. —James
Yes, you like using ChatGPT, but would you cancel your Netflix subscription to afford it? Hell no!
What else we’ve read
iOS 18 is here. Apple Intelligence? Not yet
Today marks the official release of Apple’s new flagship iOS update, but those expecting to dive into Apple Intelligence will have to wait a bit longer, as the first AI-related tools won’t arrive until iOS 18.1. Those on the developer beta can try out 18.1 now, which includes summarized notifications, a sassier Siri, and several (re)writing tools.
One major Apple Intelligence feature that is still MIA, however, is image generation. According to Bloomberg Chief Correspondent (and longtime Apple insider) Mark Gurman, both Genmoji and Image Playground won’t be introduced until iOS 18.2 due out in December, though like with 18.1, you’ll most likely get a chance to try it out sooner via the developer beta.
Earlier this year I predicted Genmoji will be a gateway drug to more generative image acceptance—a more personalized way to shitpost among friends. I still think that's the case, but that's in part because we're still a generation or two off from most folks being able to try it. Right now the cheapest entry point to Apple Intelligence is either a new iPhone 16 (starts at $799) or possibly a used iPhone 15 Pro. Next year that entry point will be $100 cheaper as Apple makes way for the iPhone 17, and so on and so forth. (Android’s system-level image generative tools have similar price points.)
Even though it’s not a part of launch, Apple is still continuing to play up Genmoji as a core Apple Intelligence feature in its iPhone 16 materials, such as its all-important “introducing iPhone 16” video and this bizarre iPhone 16 guided tour featuring comedian Sheng Wang (sidenote: don’t judge his humor for this bit and instead check out his wonderful standup special on Netflix). I would expect Image Playground to be given similar screentime for the inevitable October Mac/iPad event. All of which is to say, it continues to be clear that at this stage, Apple still wants image generation to be a key feature of the Apple Intelligence suite alongside the less visually exciting (but infinitely more useful) Siri, summary, and writing tools.
In the meantime, you'll have to settle for this AI-generated image of Craig Federighi’s dog, Bailey, which, as Wired reports, is the first Image Playground created image we've seen outside of Apple's tightly-controlled marketing materials. But of course, it also hits all the points Apple cares about. It's unambiguously AI, mitigating the risk of misinformation; and it's clearly meant for a narrower, more personalized audiences (ie. those who know the very good doggo Bailey) over a chat app like iMessage, meaning less obvious concern of the generative tool taking away work from professional artists. In other words, mostly harmless. And yet, it’s also nothing we haven’t seen before from Midjourney, DALL-E, Llama, and the like—whereas Genmoji feels like a more interesting, narrower application, Image Playground here feels like something more commonplace in the generative space, which in turn makes it feel decidedly less Apple-y. —Ross
Tutoring a Large Language Model
AI is a petulant child, both hungry for new information and confident in everything it says, no matter how utterly wrong it might be. It lacks common sense and media literacy, and because the internet has become so enshittified—thanks in no small part to AI-generated work that can cause an AI to be dumber just by reading it (no, really)—that big tech companies have found it valuable to essentially hire tutors to teach its large language models how to produce better outputs.
Writing for The Guardian, Jack Apollo George offers insight into his own time spent as an "AI annotator," ie. teaching LLMs how to better perform its own tasks and "provide gold-standard examples for the model to learn from and emulate." It's worth reading in full, if for no other reason than to understand just how much human labor continues to be involved—and how dumber AI could be if left to its own devices. I also found it a fascinating contrast to Josh Dzieza’s 2023 report for The Verge on outsourced AI annotators. —Ross
What we’ve watched
Kaos (Netflix)
Lucy Mangan in The Guardian gave Kaos a five-star review and called it a masterpiece. She’s not wrong. This retelling of the saga of Zeus, Dionysus, Prometheus, Eurydice, and Orpheus set in a modern setting is extraordinarily good. It’s funny (in a dark, weird British way), features a stellar cast, and has a beautifully written script. Oh, and at the risk of burying the lede, Kaos stars Jeff Goldblum as Zeus, the king of the Gods. I suspect Homer and Hesiod would have cast Goldblum as Zeus themselves were they around today. This show is going to win a boatload of awards; Goldblum’s sunglasses deserve an Emmy all on their own.
The Guardian’s review is better than anything I could write so I suggest you read that. But the one thing I’d like to add is to call out the remarkable Stephen Dillane, who plays Prometheus. Dillane plays the tortured best friend of Zeus (and narrator of the show) in his characteristically understated way with an added sense of quiet menace. What I find remarkable about Dillane is how his approach can be so varied depending upon the character he is playing, whether he’s the tragic, weak king Stannis Baratheon in Game of Thrones, the clever but quick-to-anger DCI Karl Roebuck in The Tunnel, or the reserved but passionate artist Graham Sutherland in The Crown. There’s a particular scene in The Crown in which Sutherland (Dillane) unveils his portrait of the octogenarian Prime Minister Winston Churchill (John Lithgow) in which Dillane delivers the line, “Age is cruel!” with a force of emotion powerful enough to crack the rock he finds himself chained to as Prometheus in Kaos. —James
Beetlejuice Beetlejuice
Tim Burton's sequel/revival doesn't quite capture the charm of the original, and if anything, it seems to play too many of the same notes just for sake of a callback. (Without giving too much away, there's a scene that uses "Day-O / The Banana Boat Song" that makes no sense even in a world where nonsense is the default position.) What's old is too often a retread, and what's new—Monica Bellucci! Willam Defoe!—never quite stick the landing. As a movie, it's a pretty good first half that starts to creak a bit under the weight of its own setup... but despite myself and everything I just wrote, I still enjoyed the hell out of the movie. The visual gags are charming and quirky in all the right ways. Every actor gives it their all, especially Michael Keaton who has not missed a beat in the decades since he last embodied the Juice. —Ross
Slow Horses (Apple TV+)
We’re only a couple of episodes into the latest season and I'm already finding myself muttering, “holy shite, this show is so damn good” (again) every few minutes. Slow Horses is easily the best show on Apple TV+, and, as is my want, I won’t be taking questions at this time. Every word in this show deserves an Emmy. So it’s gratifying that Will Smith (no, the other one, the writer) won the Emmy Outstanding Writing for a Drama Series for the episode, "Negotiating with Tigers" in this year's awards. —James
The West Wing (HBO)
For those who haven’t been paying attention, the US election is less than 50 days away on November 5th. It’ll come as no surprise to anyone reading this that we will be casting our votes for VP Kamala Harris and Governor Walz. To keep myself in a positive frame of mind over the next few weeks, I decided to start watching The West Wing again to remind myself of how a competent (albeit fictitious) Democratic administration behaves. Aaron Sorkin’s talky-talky dialog holds up remarkably well, as does the hand-held frenetic pace of the direction and the tight performance of the cast. The greatest giveaway that this show was made in 1999 is the width of the shoulders of the suits, the clunky (but awesome) computers, and the characters constantly checking their pagers.
This year marks the 25th anniversary of The West Wing. It was a delight to see the cast reunion at last night’s Emmy Awards. —James
What we’ve played
Lies of P (Round8 Studio)
As anyone reading MBH4H will know, I am a massive fan of Sekiro (currently on NG+3), Bloodborne, Elden Ring, basically every game made by FromSoftware. FromSoft’s games are, in my humble opinion, the greatest video games ever made (once again, not taking questions) and was the reason I refused to play The Lies of P, a game created by Round8 Studio in South Korea: I considered it a blatant rip off of a FromSoft Souls game and therefore assumed, very wrongly, it would be an inferior copy. Boy, did I get that wrong.
Not about Lies of P being a blatant ripoff of a Souls game; I was completely right about that. What I got wrong was thinking that was in any way a bad thing. Lies of P is Aliens to Alien, a very different but extremely good take on a familiar and much-loved theme. This is a very good video game, and I absolutely love it—I am currently about a third of the way through my first playthrough.
Lies of P has ported the gameplay style of Sekiro, blended it with Bloodborne, and set it in a Parisien/Victorian London steampunk’esque world filled with psychotic murderous puppets and Last of Us-like zombies. Why psychotic murderous puppets, you ask? The reason, dear reader, is that the protagonist of Lies of P is none other than Pinocchio, the puppet who becomes a boy with a heart of gold who mustn’t lie. Pinocchio’s challenge is to rid the world of Krat of the scourge of mechanical and carcass zombies infecting it.
The difficulty and structure of Lies of P is typical of a FromSoft game. The combat is extremely hard—a few mini-bosses and far harder to defeat than some of the main bosses—but the gameplay is a little more linear, in both level design and storytelling, than typical FromSoft. But what it lacks in those departments, it more than makes up for in terms of art and design. The puppet characters, in particular, are fantastic. I especially love the murderous clown and strange Marie Antoinette-like spider creatures in the Opera House scene and the industrial puppets in the Great Exhibition Hall. The set design itself is, at times, extraordinary.
I love and adore this game. It is a triumph, and I am a fool for not playing it sooner. —James
Astro Bot (Team Asobi)
I've seen more than my share of headlines heralding Astro Bot as the PlayStation 5's answer to Mario, which I think is both high praise and perhaps unintended damning of the 3D platforming genre as a whole; there have been few if any games that match Nintendo's flagship franchise in terms of capturing the right blend of joy, challenge, and approachability. I can honestly say Astro Bot has come the closest to anything I've played. Each level is a diorama of delightful twists, and every nook and cranny has something worth seeking out, and it loves to create large piles of things that showcase how well the PS5 can handle very busy scenes without dropping framerate.
To be sure, Astro Bot takes many cues from Nintendo, including a not-so-secret Lost Galaxy that immediately reminded me of Super Mario World’s Star World. But I’d argue those similarities are a feature more than a bug, and as YouTube essayist Adam Millard argues in an excellent video on clones, borrowing great design ideas can be a Very Good Thing. (Note that this video also spends a good amount of time on Lies of P, and is similarly complimentary of its ability to emulate the best parts of FromSoft’s formula—if there’s a theme connecting our two games this week, it’s imitation as the sincerest form of success.)
Astro Bot has literally hundreds of callbacks to PlayStation's storied history and catalog of titles, which is an extra special treat that plays on my nostalgia very adeptly—this is unapologetically a commercial for PlayStation, and like Barbie and The Lego Movie before it, Astro Bot excels by making the most of its IP playground. I just wish the game did a better job of telling me what some of the more obscure references are—and for that matter, I wish Sony would do a better job with preservation, as many of the Easter eggs come from games that are no longer playable. —Ross