🌀🗞 The FLUX Review, Ep. 188

May 8th, 2025

May 02, 2025

Episode 188 — May 1st, 2025 — Available at read.fluxcollective.org/p/188

Contributors to this issue: Erika Rice Scherpelz, Ben Mathes, Neel Mehta, Boris Smus, MK

Additional insights from: Ade Oshineye, Alex Komoroske, Chris Butler, Dart Lindsley,Dimitri Glazkov, Jasen Robillard, Jon Lebensold, Julka Almquist, Justin Quimby, Kamran Hakiman, Lisie Lillianfeld, Melanie Kahl, Robinson Eaton, Samuel Arbesman, Scott Schaffter, Spencer Pitman, Wesley Beary

We’re a ragtag band of systems thinkers who have been dedicating our early mornings to finding new lenses to help you make sense of the complex world we live in. This newsletter is a collection of patterns we’ve noticed in recent weeks.

I’ve come up with a set of rules that describe our reactions to technologies:
1. Anything that is in the world when you’re born is normal and ordinary and is just a natural part of the way the world works.
2. Anything that’s invented between when you’re fifteen and thirty-five is new and exciting and revolutionary and you can probably get a career in it.
3. Anything invented after you're thirty-five is against the natural order of things.
— Douglas Adams, The Salmon of Doubt

👀💰 The cost of being seen

Angelyne became an LA icon primarily by erecting billboards featuring… herself. This example provides a masterclass in manufacturing fame through paid visibility. This same drive manifests when aspiring influencers buy fake followers. It appears when individuals and businesses invest in SEO services. These tactics, whether flamboyantly physical or algorithmically subtle, point towards the commodification of recognition. Presence and perceived importance are increasingly transactional. These moves are another flavor of conspicuous consumption, where the commodity consumed is attention and validation.

Purchased validation can be a primal scream. A billboard in LA shouts, “I Have the Resources and the Gall To Take Up This Much Space in Your Visual Field.” It’s a pure status display. You’re meant to see it and think, “Wow, they must be someone,”... even if it’s just Someone Who Could Afford a Billboard.

Purchased validation can be a subtle boost. SEO for your name shows off to people who are already interested enough to Google you. It’s about stage-managing that first impression, ensuring the search results convey the right message. It's like tidying up your house when guests are coming over.

Purchased validation can be a play for momentum. Buying followers on social media attempts to simulate social status. By purchasing the appearance of being popular, influential, or well-connected, those who come along later may assume an account has more value. It’s building on the idea that a crowd draws a crowd… even if the initial crowd is paid actors.

As visibility shifts to the outputs of probabilistic models, this trend may point toward a future where people pay for recognition within the ‘minds’ of AI systems. While we cannot currently pay LLM providers to push their chats to learn about us, history tells us that, regardless of the medium, people will pay to be found.

What psychological service might come from validation purchased through an LLM? Today, AI models are influenced by the content they are trained on. Influence is indirect. A sort of “proof of work.” Showing up in the output of an LLM signals prominence, “Oh yeah, the AI knows about my work.” It’s less gaudy than a billboard but potentially signals a deeper, more intellectually validated form of importance.

Ultimately, all of these methods serve as means of signaling status to others. We project images onto the social screen, using money as the projector lamp. These are shortcuts to the social signifiers that would otherwise require time, effort, achievement, or genuine connection… if they are accepted as authentic.

But as our lives are further dominated by purchased validation, the dynamics of the system will change. Does purchased social capital depreciate more quickly than earned social capital? Does all social capital depreciate faster when enough is purchased? Will we develop ethical guidelines around transparency of purchased validation? How will this damage the perceived objectivity of these systems? And if we start seeing these mechanisms in LLMs, what can we learn from the historical rhymes with the mediums of the past, newspapers, radio, television, the web?

🛣️🚩 Signposts

Clues that point to where our changing world might lead us.

🚏🛴 Kickstarter is launching a “tariff manager” for creators

The crowdfunding platform Kickstarter is unveiling a tariff manager product for its US-based inventors. Tariffs enter the picture during the fulfillment phase; backers who provide sufficient funding to a project are typically promised a copy of the product when it becomes available. Inventors will now be able to add a surcharge for tariff fees, meaning that backers will have to pay this extra fee if they want their promised product. It could become a sticky situation because the backers (naturally) didn’t agree to this fee when they pledged money, so some might want a refund from the creator, who has probably already spent the money on product development. (Of course, the alternative — the creator swallowing the tariff fees — isn’t ideal either.)

🚏😵‍💫 20% of AI-suggested code libraries were fake, leading to “slop squatting” attacks

LLMs are known to hallucinate the names of code packages, confidently stating that you can import code from some library that doesn’t exist. In one test, an average of 20% of the packages suggested by LLMs were fake; commercial models did better (GPT-4 only had 5% fakes), while open-source models did worse, averaging a 22% fake rate. Usually, such AI-generated code will simply fail to build or run. But cyber-attackers have started “slop squatting”: they’ll publish malicious modules under common hallucinated names, thus becoming able to “steal data, plant backdoors, and carry out other nefarious actions” in unsuspecting programmers’ code.

🚏🚙 Subaru Canada is shifting manufacturing from the US to Japan to avoid tariffs

The North American auto manufacturing industry is famously cross-border, with intermediate goods and finished cars jumping between the US, Canada, and Mexico multiple times. But the US’s new tariffs may be rewiring that arrangement. One notable example is that Subaru, which doesn’t have production facilities in Canada, has historically sourced 26% of the cars it sells in Canada from the US. For the 2026 model year, though, that figure will fall to just 10%, as Subaru plans to shift Canada-bound car production away from its plant in Indiana and instead import finished cars directly from Japan, thus sidestepping the US tariffs.

🚏🐻 California passed Japan to become the world’s fourth-largest economy

Californians have long bragged that, if they were their own country, they’d be the fifth-largest economy in the world, behind the US, China, Germany, and Japan. Thanks to a 6% GDP growth rate last year, the Golden State has now surpassed Japan, with a GDP of $4.10 trillion compared to Japan’s $4.02 trillion. (Germany, at $4.65 trillion, isn’t far off.) California’s GDP growth indeed surpassed that of the US overall (5.3%) as well as China (2.6%) and Germany (2.9%).

📖⏳ Worth your time

Some especially insightful pieces we’ve read, watched, and listened to recently.

Shipping Time and Delays on Tariff Feedback (Molson Hart via Max Boot) — The head of a consumer products company warns that Americans won’t feel the effects of tariffs until mid-May, since it takes over a month for ships from China to make it to the US. The corollary is that, even if we were to turn off tariffs today, it would take over a month for things to return to normal. Describing how the delay in this feedback loop makes it hard to change course, the author quips that “it’s almost like we’re speeding toward a brick wall but the driver… doesn’t see it yet… by the time he does, it’ll be too late to hit the brakes.”
Are “AI” Systems Really Tools? (Jürgen Geuter) — Reflects on the concept of tools, arguing that tools are specifically designed to solve a specific task and are opinionated, guiding you toward one correct solution; they thus embed knowledge into an artifact. Most current LLMs aren’t designed to do any particular thing, so while they can be helpful at coding or other tasks, they’re better thought of as “makeshifts”: objects that you can repurpose into doing unintended tasks (like using a screwdriver to open a bottle of beer) but that don’t encode domain knowledge or communicate solutions.
AI’s Impact on the Written Word Is Vastly Overstated (Byrne Hobart / The Diff) — Argues that we are already sifting through an effectively unlimited supply of human-written content. AI-generated articles would need to be of the very highest quality to compete with this existing pool, and that is unlikely. (One weakness with this argument is that it can be challenging to find exactly the article you want, either because it is impossible to find or because it does not exist. An AI-generated article on such a long-tail topic would therefore have no competition from human writings.)
Can o3 Beat a Geoguessr Master? (Sam Patterson) — Examines how GPT’s o3 reasoning model was able to make shockingly accurate guesses on Geoguessr, a game where you’re plopped somewhere on Google Street View and challenged to find your location on a map. O3 did indeed narrowly beat the author in a five-round game (even with the web search feature disabled!), but the difference in methods was interesting: while humans rely more on heuristics, o3 has an encyclopedic knowledge of geology, architecture, road sign design, and traffic patterns in even tiny towns. Still, o3 lost some rounds, and it was much slower than a human.

🔍🔺 Lens of the week

Introducing new ways to see the world and new tools to add to your mental arsenal.

This week’s lens: the innovator’s… trilemma?

It’s time for a complete refresh of your product’s UI. You build something beautiful: fast, responsive, crafted with modern frameworks and best practices. But it lacks key features and breaks compatibility with critical tooling. Still, the team is fully committed to the new version. The problem? The old version still underwrites the paychecks, and it will for a while.

The innovator’s dilemma is familiar: incumbents struggle to embrace disruptive innovations because doing so cannibalizes their current business, but ignoring the innovation creates the risk that they’ll be overtaken.

However, in systems where the incumbent solution feels expendable, a third dynamic emerges—what we might call the innovator’s trilemma. This third branch reveals a paradox: if belief in the new solution crystallizes too early, the existing system may be neglected prematurely.

This isn't just harmful for the legacy system, which may start to crumble under the neglect. It creates a hidden fragility for the new. Like a child tragically left to their own devices, the emergent innovation is thrust into maturity before it's ready. What should be a phase of exploration and iterative learning becomes burdened by immediate expectations and critical dependencies. The old thing is deprecated, but the new thing isn’t ready yet.

Premature belief can be just as dangerous as delayed action. Innovation within an existing system needs scaffolding—support for the old even as the new emerges. Don’t let symbolic readiness outpace operational readiness.

🌀🗞 The FLUX Review