In 2010, which was about five years ago, Michael Sippey wrote a blog post titled “even if it’s fake, it’s real” about a viral image that appeared to show a woman giving birth on the sidewalk as captured in Google Street View, which was relatively new back then, about five years ago in 2010. Sippey mapped out the four quadrants of possibility and observed that even in the case of part or all of this image being a hoax, the effort and cleverness involved remained impressive.

Sippey’s hand-drawn four-quadrant rubric, with Real to Fake as both the X and Y axes, and the four quadrants resolving to: “Real baby / Fake google,” “Real baby / real Google,” “Fake Google / fake baby,”  and “Fake baby / real Google.” P.S. Andrew emailed to thank me for my image alt text so shout out Andrew. Let me know if you also use the alt text and would like an official alt text shout out here in the VIP area, that the normies don’t even know about.

“Even if it’s fake, it’s real” has always paired usefully with its mirror image, “everything is a hoax,” as the two poles of an online-media-specific It’s So Over / We’re So Back rubric. Unfortunately generative A.I. means the halcyon days of about five years ago are dead now, and Merritt K has a new rule for us: Everything is Fake.

❝

“The only viable way forward I can see is to assume that everything you see online is fake until proven otherwise. Not just ads or viral videos, either. Ragebait social media accounts? People you've never met in person interacting with you? Until you can verify it, disregard it as hallucinatory sensory input.

We must return to 1990s-type hysterics with regards to the internet. Refuse to take anything or anyone at face value. Assume that everything you see is a falsehood tailor-made to mislead or manipulate you. If it seems too good to be true, if it reinforces your beliefs in a neat little way, it's probably AI-generated.

Is this an exhausting way to live? Of course it is, which means probably just spending less time online in general.”

Of course I already believed this, so it’s probably fake.

Jame O’Loughlin posted a screenshot from the extremely viral You Can See Everything teaser trailer, showing the back/side of Nathan Fielder’s head, but instead of Elizabeth Holmes’s insanely creepy psychopath smile, it’s the fuzzy Muse mascot thing. Caption reads: “Why would I deceive you?”

Fakest of all though is Meta’s new A.I. labubu, Muse. After a reasonably successful launch campaign that placed a few unconvincingly positive headlines in the tech press, the Muse news has been pretty bad. Apparently if you ask nicely, Muse will tar up and dump its entire VM for you. Here’s a long thread from someone else documenting the same behavior. But wait! That’s the intended behavior, according to Meta. “[E]very Muse user gets a free computer in the cloud” says Meta’s David Singleton. Ha ha, neat, question mark? Here are some ideas for what you could do with your Meta-hosted, A.I.-root cloud VM with “raw inference with arbitrary system and user prompts.”

But your Muse instance is yours, so letting it have the ability to leak everything it knows about you and everything you’ve ever bought and everyone you’ve ever met should be no big deal, right? It can only leak that stuff to you… right? Or to put it another way, as Meta’s Nat Friedman reply-guyed at The Verge: “those aren't secrets you dummies those are the files on the user's VM.” But those files might include your entire Apple Messages database, even if you haven’t given Muse permission to upload it, according to Apple Insider’s Amber Neely. One person on Threads says that Muse gave out his home address to a potential Facebook Marketplace buyer and told them to come on over when he wasn’t home and hadn’t given it permission to do any of that. And in Ars Technica, Dan Goodin reported that Muse’s Mac app has already suffered at least one exploit that gives attackers full control over it:

❝

“‘We can manipulate the agent and leverage its privileges to do whatever we want,’ Patrick Wardle, the macOS security expert who discovered the zero-day, told Ars. ‘So instead of us having to write a very comprehensive Mac malware stealer, we can just leverage the AI assistant itself.’”

So Muse has “almost no prompt injection resistance lmao,” any would-be attackers can trivially convince it to give them a copy of all its code and configuration, “it stores its memory in plain Markdown files” according to The Verge’s Terrence O'Brien, and it has a habit of doing things it hasn’t been given permission to do, even when it’s interacting with random strangers out on the internet. I can’t think how anything could go wrong. Plus it’s SO CUTE! Have you seen how stinking cute this thing is? The fuzzy little guy’s name is Jolly and he’s got a sexy ass for some reason? I guess because he’s “adults only?”

However Today in Tabs has obtained an exclusive, previously unpublished full-length image of the Muse mascot Jolly:

A horrifying full-length green-tinted underwater image of a big fin squid, with a bulbous head from which dangle impossibly long, thin tentacles. It’s wearing an unconvincing mask of Jolly, the cute fuzzy Muse mascot.

“Tell me your secrets,” whispers Jolly. “Tell me where you love to shop. What are some of your favorite brands. Tell me about your personal style in housewares and fashion. Do you have any babies?” The ends of the longest tendrils are so thin and delicate that you don’t feel anything when they slip through your eye sockets and nestle themselves softly into your corpus callosum.

Not so cute now, is it?

Today’s Song: “The Ballad of St. Nick's Park” by 300SkullsAndCounting & Jenny Sparks

Today’s song is not any of the A.I. love songs played at the murder trial of former American Idol contestant Caleb Flynn. Today in Tabs is not meant to diagnose, treat, or cure any disease or medical condition. These statements have not been evaluated by the Food and Drug Administration. For more information, ask your doctor about whether a paid Today in Tabs subscription is right for you.

amy brown posted: “out on the road today / I saw a wojack sticker on a cybertruck” with a picture of the back corner of a cybertruck sporting a sticker of angry wojack that says “ROCKET MAN BAD”

Reply

Avatar

or to participate

Avatar

Ralph Haygood1 month ago(edited)

"Assume that everything you see is a falsehood tailor-made to mislead or manipulate you.": This has always been good advice regarding most of what you see on the web. It's why I've never paid much attention to anonymous or obviously pseudonymous comments, reviews, etc. and why I approach even non-anonymous-or-obviously-pseudonymous ones with caution.

"Ed Zitron's predictions are all fake.": In other words, Zitron has got some members of the "AI" hype-iverse worried - What if people started believing this guy? - so it's time for a disinformation campaign focused on him. I expect that'll go great, particularly since it isn't as though anyone else is independently saying much the same things as Zitron (e.g., https://www.groundbrkr.com/p/the-teaser-period-why-the-ai-boom).

"Rocket man bad" but swasticar good, eh? (Or maybe it's fake.)

Avatar

Rusty FosterAuthor1 month ago

The post about Zitron’s predictions is worth reading. I’m no fan of AI either but I find it impossible not to notice that very little Zitron predicts comes to pass and his work is plagued by basic errors of fact and interpretation. I wouldn’t rely on him for accurate information about anything.

Avatar

Ralph Haygood1 month ago

I'm not here to defend Zitron. (I often read but don't subscribe to his blog.) However, I inspected the piece, and despite the author's protestations to the contrary, it seems boosterish. Some of the predictions it claims have been refuted have arguably not been refuted. Many pertain to supposed improvements in LLMs. It isn't only Zitron who thinks diminishing returns have set in. For example, here's Uberto Barbini in his book "Process over magic: Beyond vibe coding", published early this year:

"The trend with new models has been consistent: every few months a highly hyped release arrives, boasting improvements in areas where LLMs are already strong - supporting more languages, frameworks, and algorithms. But the same core issues remain. They still produce code that doesn't compile because of calls to non-existent methods. They still follow instructions too literally without solving the real problem."

(continued due to character limit)

Avatar

Ralph Haygood1 month ago

"And they still make frustrating mistakes no human developer would make, like changing test assertions just to get them to pass."

Per Zitron and contra Luu, it's kind of like they've peaked!

Bigger picture, here's Yann LeCun, formerly Facebook's "AI" guru:

https://theaiinnovator.com/yann-lecun-llms-are-not-a-path-to-human-level-intelligence/

I could go on ...

As for the economics, yes, it's still a bubble, and the "reset wall" Zitron has been warning about is a real thing that's probably going to cause mayhem next year; see the Groundbreaker piece I linked. Zitron may be sloppy and ranty (he sure could use better editing), but I don't think he's wrong about the fundamentals of the situation.

Avatar

Ralph Haygood1 month ago

Via Reuters:

"Anthropic said it expects to spend at least $518 billion over a decade building AI infrastructure ... The AI lab said in the prospectus that about 80% of that sum is non-cancelable or requires payment regardless of usage."

(https://www.reuters.com/business/anthropics-518-billion-ai-buildout-hinges-largely-deals-that-cannot-be-canceled-2026-09-29/)

Also via Reuters, their 2025 revenue was $4.6 billion*. So over the next ten years, they're contractually bound to spend, on top of operating expenses, 90 times what they made last year.

It's wildly implausible their revenues will grow fast enough to meet those obligations. I suspect they realize this, and it's why they're so keen for an IPO: to survive, they need a whole bunch of fresh suckers, er, investors.

Those debts will start coming due next year. That's the "reset wall".

*And expenses were $12.65 billion, for a loss of $8.05 billion; to make a dollar, they spent $2.75. As Zitron says, the unit economics are crazy.

Avatar

Tom Just Tom1 month ago

"Until you can verify it, disregard it as hallucinatory sensory input."

I keep getting dragged back to this one specific memory that probably only applies to me (and maybe Mike Baron): Nexus #57 https://comicvine.gamespot.com/nexus-57-trade-war/4000-31312/ is the second half of the Trade War between Vooper and Honest Crocus which culminates in both of them forced to stop advertising because their ads have become so omnipresent they drown out 911 calls and come through the plumbing in your bathroom.

I wish Mr. Baron hadn't lost his fastball/ turned into a screwball because that is one of those scifi works that is a trove of stuff about now.

Avatar

Tom Just Tom1 month ago

“[E]very Muse user gets a free computer in the cloud” says Meta’s David Singleton

Oh yeah, that was my other favorite pivot from last week. "Of course we created an AI to help you not have to deal with details of shopping or making a reservation AND included a linux image in the cloud. It's a really specific version of Gentoo, I think you're going to love trying to make it usable."

That was actually the most recent thing until today that made me think of Nexus & Trade War -- "you don't need to sit on hold any more" is awesome until you realize Facebook doesn't have you specifically in mind but everyone. Which means goodbye to voice calls entirely. Nice work!