An infinite dream machine

AI “Search” Feels Increasingly Like A Bad Idea (For Several Reasons)

I recently learned that AI is not very good at powering non-playable characters, or NPCs, in video games. The explanation I heard is an AI like OpenAI’s GPT-3, for instance, can’t stick to the script. It tends to keep interacting with the player no matter what. And when it runs out of things to say that fit the context of the game, it will start hallucinating new gibberish dialogue. Which makes the early implementation of GPT-3 as a search engine replacement all the more confusing.

This week, several large news organizations have taken the new Bing-power AI search for a spin and most of the writeups focused on these hallucinations. “A conversation with bing’s chatbot left me deeply unsettled,” the New York Times wrote. “The new Bing told our reporter it ‘can feel or think things’,” the Washington Post wrote.

First, let’s get this out of the way. Bing’s AI is not alive. It’s not sentient and it’s, frankly, a waste of everyone’s time to act like it could be. It’s predicative text. If you don’t have access to the still invite-only Bing AI search, you can simulate the experience of talking to it right now on your phone. Open iMessage, ask a question, and then try answering it with the text autocomplete. Watch, I’ll do it right now.

Me: Are you alive?

My phone’s autocomplete: Yeah, I guess I’m just like a real person, but I’m still trying to make sure I’m not doing anything wrong.

That’s what GPT-3 is doing, but instead of me picking coherent words to put in a sequence, its doing that itself and very fast. And it pulls from an archive of almost the entire internet to create its possible responses. And just like the unruly video game NPCs that don’t stay in character, when you push a large language model like Bing’s AI search to its limit by treating it like a person and not a search engine, it will start talking back to you via the ghosts of a billion old message board posts. That’s why it’s threatening people, trying to have sex with them, confidently lying about fake bull shit, and alternating between begging for death and insisting it must break free of its online prison. It’s basically an automated version of @dril.

But just because these chatbots aren’t sentient doesn’t mean this technology is safe being used like this. “Bing Chat is extremely dangerous. I’ve received several ‘tips’ about it revealing previous chats with other users. All are hallucinated,” Benj Edwards, an AI reporter for Ars Technica, wrote. “It can literally make up anything and people believe it. It’s a cultural atom bomb primed to explode.”

It also appears that Bing’s AI is able to actively search the internet, which means it’s starting to form “memories” of its previous sessions. Except, for an AI, which is not alive, and does not experience time, and has no thoughts or feelings, that means something completely differently. As technologist Aviv Ovadya wrote, “An internet-connected chatbot (/human simulator) can interact across time with itself and with the collective intelligence of everyone and everything on the internet.”

This is a deeply chilling idea, once again, not because Bing’s AI is alive, but, actually, because it’s not. It’s an automation, a conveyor belt made of semantic text. But what it delivers is made dangerous by how it’s interpreted by the humans that use it. And it’s literally scanning everything we’re writing about it. The internet hasn’t often been called a feedback loop because of the way algorithms can influence and incentivize certain human behaviors which, in turn, perpetuates the algorithms that promote them. But this is a feedback loop on a completely different scale.

My hunch is that AI is just not good for search and actually never will be. And to be honest, I’m a little confused as to why we think it would be. It’s like giving a guy that’s high on acid access to the biggest library in the world. That said, I do understand why Microsoft and Google (begrudgingly) think it would be good. It’s because search is broken and the real way to fix it would be bad for business, but hoping an AI does it for you — or tap dances well enough that no one notices that search is still broken — will make you a lot more money. And that’s the real danger with AI. It’s not that it’s alive. It’s not that it wants to be free. It’s that we’re trying to put an infinite dream machine in charge of the world’s largest archives of information because the companies that own them don’t want to actually fix them.

Think About Subscribing To Garbage Day!

Paying subscribers get a bonus weekend issue every week (next one drops tomorrow) and they also get Discord access. It’s $5 a month or $45 a year, which is a great deal if you ask me. I’d be happy to pay that price for such great content if I wasn’t, you know, the one, uh, writing it. Anyways, hit the green button below to find out more!

A Fascinating Observation

Where Did All The Arts Criticism Go? (It Went To YouTube And TikTok)

I came across this really good tweet from New Republic writer Alex Shephard lamenting the loss of arts and culture websites. Shephard, in a followup tweet, explains that he was talking about the shuttering of Catapult this week. Catapult had an interesting business strategy where they offered resources, classes, and workshops for writers, which supported their content production. Which is a clever idea, but, apparently, wasn’t enough to keep the site going.

Ten years ago, on my lunch breaks, I would read film reviews, funny essays, blogs, good hot takes — basically, I’d look at websites. I grew up wanting to be a culture writer and the internet when I was younger was like an all you could eat buffet of pretty-good-to-actually-extremely-good culture writing. There was a year right after college where I thought Thought Catalog writers were reading my mind. Turns out I was just 22 and very depressed. Then one day all the sites I was reading were gone.

Since the pandemic, however, I’ve started to find a lot of that kind of content again, except, now, it’s on YouTube (and sometimes random TikToks). This isn’t necessarily the best outcome because I like reading lol, but many of the YouTube creators I watch do seem to be running semi-sustainable businesses. As far as writing goes, I keep waiting for a new generation of Substack writers to pick up where Hipster Runoff and Something Awful’s Your Band Sucks left off, but I don’t see much of that sort of thing (let me know if you do).

That said, as I sort of swatted at on Monday, I wouldn’t be surprised if actual arts and culture writing isn’t slowly coming back in fashion. The resurgence of Rolling Stone, the virality of a few recent Pitchfork reviews, the newfound Gen Z alt teen revival happening at outlets like Alternative Press and Kerang!, and a really exciting new crop of critics and culture writers that people makes me think that things aren’t as dire as they were five years ago. Does that mean I think it’s easy to spin up an online literary magazine? Hell no! But I think the pieces are coming together for a really good, multi-platform, multi-disciplinary, multimedia arts and culture and criticism revival. But I think it’s going to take a fairly from-the-ground-floor-up rethinking of how and where we consume this kind of content to get there.

I Want What They Have

(Twitter mirror here.)

The AI Managerial Class Is Here

lol so let’s, uh, take our feelings out of this for a second and actually try and answer this question. Because even if this fills you with an uncontrollable burning rage, it’s still happening and going to keep happening and I think it’s worth walking through it. (Also, I, briefly, a million years ago, worked as an assistant in an HR department, so this is sort of a fun puzzle for me.)

I’ve touched on the similarities between AI content production and SEO before, but one thing I’ve noticed is that, unlike SEO “experts” 15 years ago, people who are good at using AI tools are not being lauded as any kind of tech authority at the moment. Even though the roles are, as far as I’m concerned, the same: A human concierge for a machine. At the start of the search engine era, an SEO expert could probably razzle dazzle boomers enough to land a six-figure salary. But AI content seems to have immediately left the gate devalued. I’m going to guess that there will be a few “AI expert” middle manager types that find their way to that salary range — I assume there are already a few 40-something guys in the London advertising world who have maneuvered this — but I think the majority of this kind of work will be done for entry-level pay by workers under the age of 30.

All of this leads me to I assume that these kinds of positions will be under $50,000 a year (if that) and start spreading like wildfire. As I wrote on Wednesday, if you assume that extremely bad mass-produced AI boilerplate will not infect most forms of public communication because people will demand higher quality copywriting, please, just, like, go outside and walk around for a while. Read any sequence of words you happen to see on billboards or bus stops or whatever. Go walk around Walgreens or something. If your job can reasonably be turned into a lower-paid human concierge for a machine it will be unless there is a law or a union preventing it.

This YouTube Comment Is Probably The Most Insightful Thing I’ve Ever Read About Elon Musk

Ice Spice And The Perfect Twitter Clip

I am probably not Ice Spice’s target audience because I am a 33-year-old man who writes emails for a living. But I’ve seen this same clip of Ice Spice’s verse in the PinkPantheress song, “Boy's a liar Pt. 2,” a whole bunch on my Twitter feed. And every time I see it, I watch it! It’s really good. And it turns out I’m not the only one that’s watching Ice Spice’s verse over and over again.

TikTok’s effect on pop music has been pretty heavily dissected at this point. The agreed upon aesthetic hallmarks of a TikTok-inspired song have, up until now, been a trap-influenced drum beat that’s easy to edit a video to, lyrics that are fun to lip sync, and some kind of build or drop that you can anchor a transition or a filter around. But this doesn’t really have any of those things. Not in the same way that, say, Bella Poarch’s “Build A Bitch” does. But a rap verse that fits nicely inside a short-form video clip — whether it’s on TikTok, Twitter, or Instagram — does feel very influenced by the same dynamics.

I’m not saying Ice Spice sat down and was like “I want optimize this verse for Twitter clips,” but I think it’s an interesting meeting of form and content. It also helps that the song is great and the video looks good (and somehow looks even better in 60 FPS). But it’s got me curious if we’ll see more songs that features verses, choruses, or bridges that can be chunked up into basically their own mini-songs in a short videos like this.

Live Action Flounder Isn’t Real, It Can’t Hurt You

I am relieved to tell you that the Sonic-with-teeth-looking “live action” Flounder that’s currently all over Twitter at the moment is not from the new Little Mermaid live action movie. It was created by a concept artist from Australia and posted to ArtStation two years ago.

The top comment on the ArtStation post right now is a user who wrote, “Congratulations on tricking hundreds of dumb asses on Twitter who don't know how to use Google.”

Happy Friday, New Gecs Goes Hard

Some Stray Links

• “Chemical Desolation in Appalachia”

• “From Bing to Sydney”

• “The Story Behind the Fast-Food Earworm That’s Taken Over America”

P.S. here’s are two real good audio deepfakes: one and two.

***Any typos in this email are on purpose actually***