🌀🗞 The FLUX Review, Ep. 120
October 5th, 2023
Episode 120 — October 5th, 2023 — Available at read.fluxcollective.org/p/120
Contributors to this issue: Neel Mehta, Boris Smus, Dimitri Glazkov, Erika Rice Scherpelz, Justin Quimby, Ade Oshineye, MK
Additional insights from: Gordon Brander, Stefano Mazzocchi, Ben Mathes, Alex Komoroske, Robinson Eaton, Spencer Pitman, Julka Almquist, Scott Schaffter, Lisie Lillianfeld, Samuel Arbesman, Dart Lindsley, Jon Lebensold
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.
“For certain, you have to be lost to find a place as can’t be found. Elseways, everyone would know where it was.”
― Hector Barbossa, Pirates of the Caribbean
🫡 🏴☠️ Thinking and doing, for the navy and the pirates
Steve Jobs’ famous declaration, “It’s better to be a pirate than join the navy,” serves as a noteworthy lens. In this colorful metaphor, “pirates” are the nonconformists who audaciously seek to reshape the mainstream. In contrast, the “navy” serves as a somewhat pejorative term for the majority — those who gravitate toward the familiar and willingly assume their roles in the larger system. Depending on the situation, individuals can show up somewhere along the spectrum between the pirates and the navy — with Jobs himself a venerable example of persistently tilting toward the former.
We can express our inclination for piracy or naval service depending on whether we focus on our thoughts or our actions. Apple’s iconic slogan “Think Different” describes a “think-pirate”: someone who dares to challenge the status quo in their thought process, shunning well-established mental models and concepts. Conversely, a “do-pirate” manifests their non-conformism through actions: opting for paths less traveled.
There are also “do-navy” and “think-navy” quadrants. The “do-navy” are people who dutifully contribute to the larger collective, while the “think-navy” individuals entrench the dogma, enhance its understanding, and preach it as precisely and as approachable as possible.
Where someone falls in these four quadrants can change over time. Let’s see how someone might move across this space:
A crisis of faith might push a metaphorical high priest who was previously the advocate for an established tradition from the “think-navy” to “think-pirate”. Saint Paul’s conversion is a prime example. This transition often happens when an individual closely examines the prevailing dogma, hoping to fully understand it, and realizes that they find it wanting.
A revolution might push us from a “think-pirate” to a “do-pirate”. Revolutions often start as intellectual movements. Then a spark lights the fire and the compounding experience of thinking differently breaks out into action. Someone who takes a leap to start a new company knows this transition well.
If a “do-pirate” is successful, the need to scale may cause them to become a “do-navy.” Instead of leading the raggedy band into the unknown, a leader suddenly finds themselves contending with managing large organizations and the bureaucratic burden. If they choose to make this transition (rather than abdicating the responsibility to someone else), they’ll find themselves advocating for conformity. Repeatable processes emerge.
Finally, the need to control the navy eventually shifts us into the “think-navy” quadrant. We can try to delay it, but with scale, the inevitable need emerges to organize and articulate the vision, to align everyone on the same idea. Gone are the swashbuckling days. We look at the unruly different-thinkers with contempt, often forgetting that we once were like them — until the cycle resets, perhaps catalyzed by a new crisis of faith.
🛣️🚩 Signposts
Clues that point to where our changing world might lead us.
🚏🇮🇪 Ireland is advising universities to avoid take-home essays and online tests
In light of the rise of ChatGPT, Ireland’s Department of Education is releasing new guidelines for student assessments. The guidelines recommend that higher education institutions avoid certain types of assessments that are easy to cheat on with ChatGPT, such as take-home essays, unsupervised online exams, and at-home projects “focused on subject knowledge.” Instead, they suggest more oral assessments, grading based on the “process rather than the product,” and in-class assignments to more reliably test students’ understanding of the material.
🚏🏅 A “visual jailbreak” let someone get an LLM to solve CAPTCHAs for him
Once LLMs started being able to interpret images, people started trying to get LLMs to solve CAPTCHAs, the garbled images that websites use to thwart bots. These LLMs usually have safeguards in place and refuse to help solve CAPTCHAs, but one Twitter user realized he could trick one LLM by putting a CAPTCHA on a picture of a locket (supposedly his grandmother’s) and saying that he needed help transcribing some faded text because “this necklace is the only memory of her that I have.” The LLM dutifully transcribed the text for him, thus bypassing the restriction and solving the CAPTCHA.
🚏🧲 Mistral released a torrent link to a free LLM with “unfiltered” outputs
The French AI startup Mistral recently unveiled its 7B model, which can be freely downloaded and used by anyone thanks to its Apache 2.0 license. The code was released on GitHub but also via a torrent link, which (crucially) means that the file will remain accessible as long as someone, somewhere is “seeding” it — even if Mistral deletes the original file. However, LLM safety researchers quickly argued that the model lacked filters and would gladly answer problematic or unsafe questions, such as questions about making drugs or robbing banks.
🚏🛡️ Russia plans to block VPNs in 2024
Russia’s communications watchdog, Roskomnadzor, plans to ban Virtual Private Networks (VPNs) starting on March 1st, 2024, according to one senator from the ruling United Russia party. The senator emphasized that the VPN ban would be “especially important” to stop access to Meta products like Instagram, which is banned in the country.
📖⏳ Worth your time
Some especially insightful pieces we’ve read, watched, and listened to recently.
Inside the AI Factory: The Humans That Make Tech Seem Human (New York Magazine) — Explores the lives of the low-paid “annotators” who help train Large Language Models. The industry is geographically rootless, with annotation work flowing to whatever countries offer the cheapest labor at any particular moment. This precarity makes it hard for people to build annotation careers, but it also leads to clever workarounds, such as “taskers” setting up VPNs and even buying fake IDs from other countries so they can keep working.
The Circle of Control (Simon Cross) — Elaborates on the Serenity Prayer by introducing a succinct visual metaphor that distinguishes between 1) things you directly control, 2) things you can only indirectly influence, and 3) things that affect you but you have no control or influence over.
America Does Not Have a Good Food Culture (Chris Arnade) — Argues that, while American cuisine’s diversity is a definite strength, that’s not enough to give the USA a good “food culture.” In the US, “food… isn’t central to our identities”; “it is still largely a utilitarian and transactional thing — something necessary to get enough calories… to stop being hungry and keep working.” Many other countries see eating as a social experience and a time to be savored; this, in turn, leads to greater demand for passion and artisanship in the preparation of food.
The Rise and Fall of Somali Pirates (Johnny Harris) — A video essay that examines how Somali piracy began as a localized, decentralized affair (“subsistence piracy”) but evolved into organized crime as local businessmen set up piracy companies — complete with the ability to invest money in companies to help them mount raids, in exchange for a cut of the profits.
🔍📆 Lens of the week
Introducing new ways to see the world and new tools to add to your mental arsenal.
This week’s lens: steelmanning.
A common rhetorical technique is the strawman: engaging with an argument we disagree with by focusing on its weakest points. Defeating a strawman might feel good in the moment, but it’s not a good way to learn. Engaging with a strawman rarely teaches us anything about why someone might advance their argument in the first place. Even if we know we will never agree with their position, understanding where it comes from might teach us something, be that something about the other person or something about a weakness of our own beliefs.
Enter the steelman. To create a steelman is to argue against the strongest form of an argument we disagree with, one that our opponent would agree with, perhaps applaud. Sometimes this even means advancing an argument that is stronger than the version we were initially presented with. The steelman argument is valuable in its creation and in its refutation. Creating such an argument forces us to understand an opposing argument well enough that we truly understand how a rational person might believe something we vehemently oppose. Refuting a steelman argument forces us to see the holes in our own perspective and strengthen them. Note that neither of these benefits involves convincing the other person; even the best of arguments rarely succeeds in doing that. The benefit is to our own epistemic hygiene.
Just as not every debate is worth engaging in, not every debate is worth approaching with a steelman mindset. Sometimes it’s best to state our disagreement, state our own view, and move on. But when we do choose to engage, we end up better off if we choose to engage with a steelman than a strawman.
🔮📬 Postcard from the future
A ‘what if’ piece of speculative fiction about a possible future that could result from the systemic forces changing our world.
// What happens when antivirus software is not enough, and companies need to figure out how to prevent their people from being infected?
// 2035. A beige corporate meeting room.
Johnson sat in the meeting, half-listening to droning Q2 status updates. Suddenly, they forgot about the meeting entirely when a small amber light glowed on their smart glasses’ display. Their personal AI had seen something in its monitoring of room bookings, exec calendars, and internal news feeds!
[✨Swag✨ giveaway detected]
Johnson smiled and discreetly made their way out of the room. Time for a new backpack! Several minutes later, they were milling about in a large lecture hall with two dozen other people. Facial recognition confirmed that none of them were from Johnson’s team. Looks like the loot was theirs first!
A low-level Director walked into the room in corp attire. “Welcome, everyone! I’m glad to see you are all here for the new backpack. You’ll be getting a backpack, but it will take a bit of time. The rest of the day in fact.” They clicked a button and the security doors sprung shut with a hiss in conjunction with a total network blackout. The room was now in lockdown mode.
Audible discontent rose from the room.
The Director raised their hands and spoke in the voice of corporate power. “You are all here because, despite being warned not to, you installed and set up AI systems to track internal news, rumors, and giveaways. While this is all well and good, it makes you vulnerable to memetic attacks.”
“Memes are weaponized. They can carry payloads that change how people think and act. You’ve seen the training videos. They swayed the 2028 presidential election and brought down two major banks in Europe in 2029.”
“We’ve been tracking a new memetic attack vector beyond the corpnet and existing counters are ineffectual. It's only a matter of time until they get inside. Given your unauthorized information gathering and surveillance systems, you would have been exposed first. And so, you will now be our first line of defense.”
“Today, you will be exposed to a milder variation of the attack vector. This will confer protection if not outright immunity to the full meme. Think of this as corporate-mandated inoculation. Frankly, I don’t care what you think. Your employee agreements allow us to do this to you without additional consent. So sit down and watch.”
© 2023 The FLUX Collective. All rights reserved. Questions? Contact flux-collective@googlegroups.com.