New manifesto without anything new dropped, controlling complex systems, Algospeak in war times, and revaluing the strike.
Welcome to Around the Web.
First, a correction: In the last issue I said that there is no oversight and zero fines in space, that wasn't entirely true. The FCC fined Dish Networks $150,000 for leaving junk in space. Shortly after I published the issue, Ordinary Things published his newest video Satellites: Crimes Against Space which explains the whole situation in way more detail.
I revamped the design of my website a bit over the last weeks. I’m quite happy with it, so you may want to read this issue on the web.
Also, pardon my French German in the title, but I don’t know a translation that captures the essence of the word Denkzwerge quite like the original.
Before we start with our regular programming, here are
First, Marc Andreessen published the world’s worst meta bad take.
It’s long. It’s badly written. It feels like Andreessen dictated it into ChatGPT while being trapped at Burning Man. It makes some absolutely wild statements without any efforts to back them up. It quotes Filippo Marinetti, notable proto-fascist and OG techno-optimist. Basically, it is the 30,000 word version of the «This is fine» meme.
If you don’t feel like reading the entire thing – and there is no reason in the world why you should feel like reading the entire thing – Ben Grosser captured the essence in his redaction poetry version.
In Why can’t our tech billionaires learn anything new? Dave Karpf pointedly analyses what feels so off about the richest persons in the world crying that the rest of us don’t applaud everything they do anymore.
The most powerful people in the world (people like Andreessen!) are optimists. And therein lies the problem: Look around. Their optimism has not helped matters much. The sort of technological optimism that Andreessen is asking for is a shield. He is insisting that we judge the tech barons based on their lofty ambitions, instead of their track records.
In an interview with the German newspaper Der SPIEGEL, Theodor W. Adorno called out those «who frantically cry over objective despair with the ‹hurrah› optimism of immediate action to make it psychologically easier for themselves». Andreessen’s manifesto is the epitome of this «hurrah» optimism.
He openly illustrates the limits of imaginations of him and those in the Valley in a way no critic ever could. AI as the force that will destroy but also save the world, capitalism as the only form of economy that can work and if it doesn’t work, we need more more more more more until if fcuking works. The destruction of society will continue until morale improves.
If that’s his future, I’m happy to fight for another one.
Elsewhere in rich people who want the world to believe they are geniuses but are just rich (and fascists): Peter Thiel snitched for the FBI.
Why, you might ask? Well, probably it feels good if someone says «Thank you, Mr. Thiel, that’s very valuable information, Mr. Thiel».
Dead and kicking: Silvio Berlusconi. Apparently, he spent his retirement late-night-shopping, spending some twenty million Euros on essentially worthless works of art.
In a rare display of decency, his successor, Giorgia Meloni, split from her partner after he made sexist remarks. I wonder if, one day, she finds out about the rest of her party. Just kidding, she didn’t become their leader by accident.
Essentially worthless, but undisturbed, Jim Jordan refused to give up on this bid to become speaker of the House, it failed, and his party ousted him. Will this election cycle be McCarthy bad? I, depressed, hope not, but at the same time hope it becomes worse. Spectacle is the fuel that keeps me alive.
Rest of World generated thousands of images using Midjourney, analysing the output for racial stereotypes. They conclude that AI reduces the world to stereotypes. Make sure to read the whole story over at Rest of World. It includes visual representations of the output, which makes the point even more convincing than some text about it.
Rest of World also reported on the way China forces students at vocational schools into data labelling for its AI industry.
To combat stereotypes, it’s often said that we need to understand the algorithms, have some kind of transparency. That’s not enough, argues Rachel Thomas. Instead, we need to have mechanisms that let us contest the power those systems wield.
These mechanisms, are ever more important as complex systems are hard to control.
Focusing on complex systems leads to several perspectives (incentive shaping, non-deployment, self-regulation, and limited aims) that are uncommon in traditional engineering, and also highlights ideas (diversification and feedback loops) that are common in engineering but not yet widely utilized in machine learning. I expect these approaches to be collectively important for controlling powerful ML systems, as well as intellectually fruitful to explore.
Facebook announced some new celebrity chatbots, which manage to feel outdated while using the hot technology of the moment. Tom Brady’s chatbot incarnation quickly insulted Colin Kaepernick. Facebook said the usual thing, that these high-profile, highly expensive features are «experimental».
It’s no surprise, that some celebrities simulated by these chatbots were shilling crypto not that long ago.
After all, we are living in the age of the grift shift.
The Grift Shift is a new paradigm of debating technologies within a society that is based a lot less on the actual realistic use cases or properties of a certain technology but a surface level fascination with technologies but even more their narratives of future deliverance. Within the Grift Shift paradigm the topics and technologies addressed are mere material for public personalities to continuously claim expertise and “thought leadership” in every cycle of the shift regardless of what specific technologies are being talked about.
Building cutting-edge models requires an immense amount of data and computing power, making it basically impossible to do it without the backing of one of the larger players in the space. How Big Tech is co-opting the rising stars of artificial intelligence explains these dynamics in more detail.
It’s also the origin of the latest paper by Meredith Whittaker and her co-authors, Open (For Business): Big Tech, Concentrated Power, and the Political Economy of Open AI. In it, they argue that the current state of generative AI makes it extremely hard, if not impossible, to build truly open systems. If reading an entire paper doesn’t fit into your schedule, Whittaker joined This Machine Kills to discuss the paper.
But who, actually, makes money? For most companies currently in the AI race, profits are a possible future, but nothing they have figured out yet. GitHub Copilot is losing $20 per user and month. This might still be a power grab, but I don’t see any real competition in the coding market anyway?
OpenAI hosts its first developer conference at the beginning of November.
The models do require electricity, but also an ever-increasing amount of water. According to researcher Shaolei Ren, ChatGPT consumes as much as 500 ml water per prompt.
But it’s not just the public imagination and electricity consumption that are taken over by the race for powerful AI models. Open Philantophy, flagship of the Effective Altruism movement, sponsored the salary of multiple advisors in the US Congress.
Google’s Phones let you create composite images out of multiple images taken shortly after another. All smiles.
You might have heard about Reinforcement Learning in connection with machine learning models, maybe seen the abbreviation «RLHF» turning up. RLHF stands for Reinforcement Learning from Human Feedback. But what is this, how is it applied in the training process, and are there alternatives? Sebastian Raschka explains.
In an interesting bit of research, Anthropic managed to build a (small) model and were able to analyse features instead of individual neurons activating in generating output. This might lead to better interpretability and control of models, if the approach can scale to the size of Large Language Models.
Truth, they say, is the first victim of war. While that’s technically untrue, truth dies pretty fast, as all parties of a war have a story to tell which might or might not align with what is actually happening.
In the ongoing Israel–Hamas war, an explosion occurred near to the Al-Ahli in the Gaza Strip. Hamas was quick to denounce Israel, saying that 500 people died in the attack.
The message spread like wildfire through social media and engagement-driven news organisations (which basically means all news orgs). Open-Source Intelligence (OSINT) researchers, like Bellingcat, and teams such as BBC Verify painstakingly constructed are more nuanced picture, and as morning dawned the supposedly bomb attack turned out to be a smaller crater and some burnt-out cars.
While some OSINT researchers do what they have done over the last years by providing well-researched facts, others are using Twitter’s engagement machine to destroy the information ecosystem, as 404 Media reports.
Instagram’s translations added the word «terrorist» to posts containing the Palestinian flag.
Users posting about the war, are – as with other topics – resorting increasingly to Algospeak, the use of a language that uses symbols, abstractions, or neologisms to evade the algorithms of the social media platforms. Side note: Facebook trains their Language Model on the posts in its platform, so it will be interesting to see wether «‘P*les+in1ans» shows up in its future generations.
I won’t comment on the larger conflict here. John Ganz’s The Trap manages to formulate my feelings pretty conclusively:
Strategy and tactics are not what’s really at issue here. At core of the worldviews in question is a belief in sheer murderousness. What both Hamas and the far right in Israel want this is to become is a war of annihilation and extermination. This is the fundamental vision of their nationalism of despair: races and peoples pitted against each other in interminable conflicts that can only be concluded with “final solutions.” Of course, a similar vision of permanent racial war underpinned Nazism and the Holocaust. I categorically refuse to be recruited to this conception of the world. And I will not be manipulated by emotional appeals and propaganda—by one side or the other—to participate in it.
I enjoyed reading Farmers only, a look at algorithmic amplification in the age of enshittification.
Looking around at the overharvested fields of digital shit, it’s hard not to ask: is the personality quiz the ouroboros of the algorithm? Is all of social media just a personality test? AITA, the Twitter hypotheticals, the personality type tutorials invite us to project ourselves into a handful of predetermined choices. To pick one of seven essences. To choose between Jay Z or $500,000. It gives the illusion of randomization, customization and personalization, but ultimately all it does is produce a quantization of who we are. It turns the mass of the self into a collective of discrete and finite individual components.
WordPress announced official support for ActivityPub, the web standard that powers Mastodon.
Gmail users can now reply to Gmail users with emojis. The rest of us will be annoyed by yet another mail. It’s not without precedent, Apple did the same when iOS users reacted to messages from non-iOS users.
Do you have what it takes to lead a Trust & Safety team at a fast-growing social media site? Trust & Safety Tycoon lets you find out about the intricacies policy decisions entail.
The German Federal Cartel Office (Bundeskartellamt) forced Google to give users in Germany better control over the data Google collects from them. Users will be able to control how Google collects and combines data from different sources.
In MIT Technology Review Stephen Ornes published an in-depth look at the past and future of encryption.
Psychology is in the midst of its replication crisis, the new editor of Psychological Science hopes to combat it.
No surprise. Songtradr laid off large swaths of Bandcamp’s staff, among them most employees who were eligible in the unionisation process and the whole bargaining unit.
Apropos of this, but also everything, it is time to revalue the strike, as Erik Baker argues in Jewish Currents.
Terrorists, according to German law enforcement: Leftists who, allegedly, sprayed pro-Antifa graffiti.
No terrorists, according to German law enforcement: Nazis hoarding firearms.
Still trying to solve the trolley dilemma? Maybe learn a new language first, researchers at the University of Chicago claim that our moral decisions are dependent on wether we think in our native or a foreign language.
Ever used WinAmp? You will love the Winamp Skin Museum.
Recipe of the Issue: roasted red pepper and tomato soup with chunky chorizo-y croutons.
That’s it for this issue. Stay sane, hug your friends, and Antifa forever.
Using Webmentions and webmention.io.