Eating Glue to Destroy the Mutants

Why I have trust issues with GenAI and my thoughts on a comic book milestone.

Author

Cameron Huppertz
June 07, 2024

AI Trust Issues 

Back when Twitter was still Twitter, I had a pretty good system for evaluating the quality of any account on the site. If I read something that seemed hyperbolic or got a reply that felt a little off, I’d click over to their account and see if I’d already blocked anyone they were following. Knowing someone chose to follow a bunch of people I didn’t ever want to interact with was a really strong signal about how they were using the platform. It was a great shortcut for helping me evaluate how serious or skeptical I needed to be when evaluating their posts. 

I ended up on TripAdvisor the other week and saw they have an AI powered trip tool. In my attempt to balance my skepticism around our current GenAI moment, my desire to fairly evaluate the trend, and my general curiosity, I started playing around with it. I actually found it somewhat useful. Their tool uses OpenAI tech and Tripadvisor review data. I already have opinions on how much I trust that review data and how it can be useful for me. Am I going to go on exactly the trip it planned for me? No, but I know how to fit the info it provided me into my existing mental model for travel planning. This tool felt like the kind of useful machine learning project we used to hear called “machine learning” before “AI” became the term du jour. 

According to Google’s AI Search Overview, Artificial General Intelligence or AGI is “a type of artificial intelligence (AI) that aims to create software with human-like intelligence and the ability to self-teach. The goal is for the software to be able to perform tasks that it is not necessarily trained or developed for.” It’s the holy grail tech the AI hype men keep telling us is coming. The ambition of the big players isn’t bound to a contained data source—like just Tripadvisor reviews—it’s a future of human-like, multi-purpose, extremely knowledgeable, super AIs. But today we have Google telling us to eat glue.

When Google spat out non-toxic glue as a pizza sauce ingredient, people tracked it back to a specific Reddit comment. We know Reddit and Google have a deal to use Reddit data for training Google AI models, and this suggestion was peculiar enough to reverse engineer its source. But what happens when the answers are just a little off? Or when it’s pulling from training data that’s less publicly known? Or when we have no insight into a model’s training data? 

How do we go about building a mental model for the trustworthiness and usefulness of these tools?

It would be one thing if Google’s GenAI paste-based culinary advice came from a side product marked as an experiment or beta program. If the company in some way had indicated you might want to take that result with a grain of salt. Instead users found it on the main search page, as part of the core Google product, with the full weight of Google’s decades of brand trust behind it. All of our habits and biases around using Google to find and evaluate information come along for the ride here unless we actively work against them. Which leaves us, and Google in a bad place. Users are now getting information of an extremely questionable quality and Google risks destroying their brand reputation with it. 

I don’t mean to just pick on Google here. I think this is a problem for any of the companies trying to build AGI (or just sell you the idea they can build it). If anyone working on this tech wants people to take it seriously, the industry needs to get serious about helping people understand and evaluate what it actually can and can’t do. The hype machine and borrowed authority of the currency AI confidence game just isn’t doing that for me right now.

Platform Updates

Instagram & Threads

It’s a grab bag of Instagram and Threads updates and test this week. On Threads, you can now swipe on a post to tell the app if you want to see more or less content like it in the future. One, I’m skeptical of how much this actually does. Two, this level of algorithm feedback feels so small and outdated compared to Bluesky’s custom feeds.

On Instagram proper, they rolled out some updates to Notes, and added more features for their Broadcast Channels. I hate Broadcast Channels but I guess some creators and fans must be using them. They are testing some new text overlay and image formatting options in the carousel post creation flow. Oh and they are testing unskippable ads. 

Don’t panic just yet, Meta tests a ton of stuff that never goes to GA. But as someone who spends a lot of time talking to big advertisers and media agencies, I can tell you there’s a lot of interest in buying this kind of ad. Is Meta the place for making that buy though? I tend to agree with Ryan Broderick’s take over at Garbage Day, “there is not a single piece of content on any Meta platform that is worth sitting through an unskippable ad to watch the entirety of.”

The Rest of Meta 

TikTok

YouTube

Google 

Twitch 

LinkedIn 

Twitter, Sorry X

I’ve been largely ignoring recent Xitter updates because they don’t really matter, but I do want to point out two things this week: 

First, Elon’s telling Nvidia to send chips ment for Tesla to X and xAI. He’s asking the supplier of an in-demand tool to redirect resources from a publicly traded company to his privately held companies. If Tesla had a real board of directors, he’d have been fired and sued for this. 

Also this week Xitter changed their rules to officially allow adult content (which I guess was just unofficially allowed up until now). This change comes at the same time the company’s removing the ability to see who has liked another user's posts. I tried to log into my account to verify how that rollout is going and um…

A screen recording of the x dot com login page. As soon as the page loads it glitches out and resets itself making it impossible to click anything or login to the website. 

Anyway I’m sure these two changes are in no way related and have nothing to do with what Ted Cruz used to like on Twitter. 

Filling Space

This picture of my cat BMO  is here to fill up space so you can scroll past that distracting glitchy X login page gif and get back peacefully reading Kinda Brief. Also BMO is very cute.

Culture Movers 

Film & TV

Music 

Publishing 

Gaming 

Godzilla

Scams

End of an Era 

This week closed out a multi-year period in X-Men comics, and I’m a little in my feels about it. The Krakoa Era started back in 2019 with Jonathan Hickman’s two interlocking miniseries House of X and Powers of X (HoXPoX to fans with character count concerns). I’ve talked about my love of X-Men and this run of comics a few times here in Kinda Brief, so you probably know I care a lot about this franchise, but this specific era of stories was really special. 

The Mutant Metaphor, how the fictional experience of marginalization relates to real world oppression and discrimination, is a central theme in X-Men comics. Protecting a world that hates and fears them, and all that. The problem with spending 60 years as a fictional universe’s stand-in for marginalized groups is that you always get reset back to being hated and feared. No matter what civil rights wins or progress real world groups achieve, there’s always another struggle to address or another group to include in the expanding Mutant Metaphor. This can sometimes make it challenging to tell new, interesting, and authentic stories with the characters. And for the fans who identify with the X-Men because they see their own fights for acceptance, dignity, and love in these characters, it’s hard never seeing them achieve any lasting progress in that fight against oppression. 

Krakoa felt different. For five years we had a radical vision of what it would look like for mutants to put some Ws on the board. In HoXPoX, mutants establish their one island nation on/in partnership with the living island Krakoa. They invent new technology using Krakoan plantlife and combinations of mutant gifts allowing them to build teleportation gateways and resurrect the dead. The stories aren’t without conflict. You’ve got internal power struggles, and external threats. But if you ever saw yourself in the little kid hiding their mutation or running from bigots in the comics, you didn’t have to imagine waiting for the X-Men to find you and draft you into their paramilitary strikeforce / school. You could just walk to the nearest Krakoan gate, and find a new home on an island paradise where you didn’t need to run or hide.

I’m a few months behind right now, but I plan on catching up to this week’s final Krakoa Era issues over the next few weeks. I’m excited to see how the mutant conflict with evil AI resolves. Oh yea, the core conflict running through all of this era of comics is about evil AI trying to destroy mutants and become a machine god. Feels relevant. 

PS

If you wanna jump into X-Men stuff, X-Men 97 season one is all out on Disney+ and is a very good jumping on point. Marvel Unlimited is a great option for getting into the comics. It has reading lists for events and larger stories like HoXPoX. 

*I’m very dyslexic, and this is a largely free project/hobby. I do not set aside the same time for proofreading that I do for other professional work. If you spot a typo that would cause a communication error, please reach out to gently let me know.