SubStack's Dilemma; & BRIEFLY NOTED: For 2023-04-15 Sa

SubStack's "nazi bar" dilemma; pick-up collective action via spontaneous democratic organization; neural networks, attention, & transformers; human progress; Farrell, Schmidt, Barro, & Hammond...

FOCUS: SubStack’s Dilemma:

Let’s turn the mic over to Ken White for a second:

That is a good summary of what happened to SubStack CEO Chris Best on Decoder <https://www.theverge.com/23681875/substack-notes-twitter-elon-musk-content-moderation-free-speech>. Mike Masnick at TechDirt has receipts: Substack CEO Chris Best Doesn’t Realize He’s Just Become The Nazi Bar:

You end up with malicious users who cause trouble. And trouble drives away users, advertisers, or both. And if you don’t deal with the malicious users, the malicious users define you. It’s the “oh shit, this is a Nazi bar now” problem. And, look, sure, in the US, you can run the Nazi bar, thanks to the 1st Amendment. But running a Nazi bar is not… standing up for free speech. It’s building your own brand as the Nazi bar and abdicating your own free speech rights of association to kick Nazis out… and… craft a different kind of community…. It was understandable a decade ago… that not everyone would understand all this. But it is unacceptable for the CEO of a social media site today to not realize this. Enter Substack CEO Chris Best…

What Chris best and Hamish McKenzie want to be is: infrastructure. They want to provide people with a software platform so that they can run their own newsletters, via email, and moderate their own communities of discourse. And they also provide you with an email address @substack.com plus a website subdomain of <http://substack.com> at which you can park the archives of your newsletter. They thus provide you with a software platform, not a social-networking platform.

(I do not know whether they would actually except a customer who wanted to park their SubStack at <http://liebstandarte-nazi-ar.substack.com>, but they are tending in that direction.)

Now, all that is well and good, defensible, and sustainable, but they want more. They say:

1. Let us hire people who we think would help give SubStack a great reputation, direct one for a year, and then, if it takes off, to continue it on their own. They thus make editorial decisions with respect to choosing what kinds of writers—with what kinds of politics and morality—they want the idea of SubStack to stand for. That is just a little bit not-a-platform-but-an-editorial-operation, but…

2. Let us add a discovery engine—ask our SubStack authors to recommend other SubStacks, and then send messages to their subscribers: “Author[|s] you follow recommend SubStacker X. Perhaps you would enjoy their work?” Note that this is not an organic recommendation of X coming from SubStacker Y whom you have invited into your email inbox. It is, rather, a SubStack Central-run crowdsourced recommendation engine matching you to SubStacks based on the graph of your follows and their recommendations.

So far, I think, so good. Well, (1) perhaps not completely so good. But a time-limited jump-start initial “we think there is a natural fit with our infrastructure platform”is understandable at launch: needs must when the Devil drives, after all. But (2) is definitely still “so far, so good”. Why is it still “so good”? Because nothing is showing up in your feed unless you have actively and personally taken steps to put it there.

But now they say: “Let us add SubStack Notes, which gives you a “Home” timeline feed from your “extended SubStack network”. What is your extended SubStack Network? It is made up of notes from:

1. writers and readers) you have subscribed to or have selected to see more from…

2. writers whose publications are recommended by SubStack authors you have subscribed to…

3. Restacks of SubStack content by and replies to note-posts from people you have subscribed to…

And here (2) is definitely trouble. SubStack is taking note-posts that you have not Subscribed to, and that people you follow have not decided on an individual basis, and shoving them into your face.

(3) is also lesser trouble. I have signed up to see author Y. I have not signed up to see author Z, whose work is being liked or amplified by author Y.

This “Home” Notes timeline strikes me as a big mistake. It transforms the situation away from one in which SubStack is providing infrastructure, and we readers are building and following individual discourse communities It transforms the situation into one in which the SubStack ALGORITHM is in control. And so Chris and Hamish are either on the road to becoming The Nazi Bar, as the Gresham’s Law of the internet takes hold, or they will need to start moderating with the heavy hand.

If I were running SubStack Notes, I would have five timelines:

1. DEFAULT: Note-posts from and replies by people you have subscribed to…

2. Extended: DEFAULT plus “Restacks” by people you have subscribed to…

3. Maximum: Extended plus likes by people you have subscribed to…

4. ALGORITHMIC Bestsellers: Highest-traffic recent note-posts and replies…

5. ALGORITHMIC Experimental: Note-posts from authors whom you have not subscribed to, but who have been recommended by authors who you have subscribed to…

And I would have you, each time you try to access one of the ALGORITHIC timeline feeds, see a trigger warning: “The timeline feed you are about to access contains note-posts and replies that have not been individually written by or directed into your feeds by authors whom you have actively subscribed to. Proceed with caution.”

And, in addition, each note-post or reply in the ALGORITHMIC timelines must come with a provenance attached, explaining what the chain of social-graph links is that has put this particular note-post in your feed, and what concrete actions by named individual humans put it there. Then you have a chance of changing the narrative from “Substack is running an algorithmic timeline that is turning this place into a twitter-like toxic sewer” to “gee: Y, who I subscribed to, is a clown for putting this into my timeline”.

Then, of course, there is the question of whether Chris, Hamish, and company actually want to get into the algorithmic-feed business at all. I understand that they very much want people to find more SubStacks that they will enjoy, that they will find inform them, and those that they will be willing to pay for. But accomplishing that, without going over the line into the outrage-engagement-clickbait destructive vicious circle will be very, very difficult. Quite possibly they should direct their energy in other directions.

Josh Barro’s recommendation that only those paying for the SubStack service infrastructure should be allowed to post or reply seems to me too have a great deal of potential wisdom in it…

---

MUST-READ: Crowd-Sourcing Collective Action:

Nick Gruen sends us to:

Tanner Greer: A School of Strength and Character: ‘The first instinct of the nineteenth-century American was to ask, “How can we make this happen?” Those raised inside the bureaucratic maze have been trained to ask a different question: “how do I get management to take my side?”… The nineteenth-century pattern of life that created it no longer exists. But the features of its social fabric demonstrate what an agentic society with a recognizably modern set of technologies and institutions looks like…. Three features are especially prominent: the aspirational ideal of public brotherhood, a commitment to formality and discipline in self-government, and organizational structures that combined decentralization with hierarchy. These are the same patterns any future culture of high agency self-government will also have to cultivate in themselves and their neighbors…. Five decades after the Albany-Boston Railroad finished construction, the Englishman James Bryce reported that the Americans still “seem to live in the future and not the present…they see the country not merely as it is, but as it will be twenty, fifty, a hundred years hence.”… [For] Alexis de Tocqueville… Yankee agency became an object of fascination…. A country whose citizens generally had no extensive kin network to rely on. Despite not sharing blood, people worked with colleagues and strangers on the basis of shared, socially enforced norms of behavior, as well as moral codes that privileged behaviors like truth-telling, honest effort, and fairness…

Prestige, reciprocity, redistribution, propaganda, hierarchy, democracy, honor, bureaucracy, market system, and perhaps now the algorithm—this is my list of the perhaps nine ideal-typical modes of human social organization. Of these moods, four—propaganda, democracy, honor, and market system—are different from the others in that they provide space for crowdsourcing: for utilizing much rather than only a tiny portion of collective human brain power to solve problems. It's 19th century pattern that de Tocqueville called “the pursuit of self-interest rightly understood”, via pick-up self-organization and collective action, seems to me to be a culturally- rather than institutionally-based form of democracy.

It would be very nice to get it back.

---

ONE VIDEO: Recurrent Neural Networks, Transformers, and Attention:

---

ONE IMAGE: Human Progress!:

---

Very Briefly Noted:

1. Gillian Tett: America must expand its friendship group in the interests of trade: ‘Restricting supply chains to trusted countries is fraught with danger…

2. Barry Eichengreen: A Bank Murder Mystery: ‘The incompetent-management view.… The incompetent-customer view…. Incompetent regulators…. Ill-advised macroeconomic policies…. The result is a bit like the conclusion of Agatha Christie’s Murder on the Orient Express: everyone did it. The question is what to do about it now…. The only viable solution is more effective bank regulation…

3. Frederik Gieschen: Warren Buffett's Lectures at the University of Notre Dame (1991)…

4. Zeynep Tufekci: Here’s Why the Science Is Clear That Masks Work…

5. Ratul Ghosh: How To Build Your Own Question Answering Chatbot for Youtube Videos…

6. Julian Togelius: Choose Your Weapon: Survival Strategies for Depressed AI Academics: ‘Are you an AI researcher at an academic institution? Are you anxious you are not coping with the current pace of AI advancements? Do you feel you have no (or very limited) access to the computational and human resources required for an AI research breakthrough? You are not alone; we feel the same way…

7. Josh Marshall: Backchannel 81: Trump’s Rolling Insurrection: ‘Trump’s message is clear: my safety and freedom from accountability is more important than the public safety and the safety of the state itself. I come before the republic and the state. Trump’s party, or at least the better part of it, seems ready to comply…

8. Paul Campos: We’ll make great pets: ‘Remember that story from way back when about how a billionaire bought Clarence Thomas’s mother’s house, fixed it up, and maybe got rid of a troublesome neighbor in the same transaction? Well it turns out that Mom still lives there…

9. Noah Smith: 2023 is when the empires strike back: ‘China is not an ideological, proselytizing power; its ideology, basically is just “China”. Xi Jinping doesn’t care whether you have elections and protect civil rights or send minorities to the death camps, as long as you support Chinese hegemony abroad. Cold War 2 is therefore a bit more like World War 1—a naked contest of national power and interests. And if the U.S. tries to turn it into an ideological battle, it could backfire…

10. Langchain: Tutorial: ChatGPT Over Your Data: ‘Setting up ChatGPT over your own data…. [1] Load data sources to text…. [2] Chunk text…. [3] Creat[e]… numerical embedding[s]…. [4] Load embeddings to vectorstore…. [5] Combine chat history and a new question into a single standalone question…. [6] Lookup relevant documents… [via] embeddings and vectorstore…. [7] Generate a response…

---

¶s:

Henry Farrell: Industrial policy and the new knowledge problem: ‘Modern industrial policy… [requires] investment and innovation decisions [that] involve tradeoffs that market actors are poorly equipped to resolve…. [Yet] we lack the kinds of expertise that we need to achieve key goals of industrial policy, or to evaluate the tradeoffs…. This lack of knowledge is in large part a perverse by-product of the success of Chicago economists’ rhetoric…. This has a variety of consequences. Policy mistakes are more likely. Market actors find it easier to manipulate the understanding of government policy makers, e.g. as to the extent and kind of subsidies required in particular sectors or for particular purposes. One way to remedy this is to rethink the kinds of specialist education that public administrators receive, both to ensure that low and mid-level functionaries are better equipped to take the decisions they need to take, and to signal increased prestige for non-traditional forms of policy knowledge…. Elite US policy schools… have by and large converged on a framework derived from a watered down version of neoclassical economics…. New skills, including but not limited to network science, material science and engineering, and use of machine learning would be one useful contribution towards solving the new knowledge problem…

---

Steve Schmidt: FDR's legacy: saving civilization: ‘A vast taker class of America’s elites has destabilized American democracy by obliterating the American dream. The crisis America faces is real, present and deeply unsettling. Our politics is broken, but so are many other things. Yet, there is no cause for despair. Instead, there should be a sense of gratitude around the opportunity to live in consequential hours when the tectonic plates of history are moving towards the end of one era and the beginning of another. The era that is ending was shaped by the political genius and far-reaching vision of a man who died at 63 years old, 78 years ago today…

---

Josh Barro: The Arrival of 'Normal Politics' for the Debt Ceiling: ‘You also see some crazy attitudes about disorder—see this other LA Times article, where it’s not just the subjects being ridiculous but also the reporter, who is very concerned that too-loud classical music might drive homeless people out of Westlake-MacArthur Park station… [and] literally describes the music as torture: “[E]levated volume, coupled with repetition, is a way that music has been used as torture throughout history, says [musicologist Lily] Hirsch. Constant exposure to loud music can disrupt sleep and thought and eventually make people lose their connection to themselves…”. The choice LA faces…. Make the system clean and orderly and acceptable to commuters, or let it devolve into a horrifying and unsafe drug den (possibly with a repetitive soundtrack). One reason the city has so far chosen the latter option is that a substantial minority of Angelenos seem to think it’s the desirable one…

---

Samuel Hammond: OpenAI’s big lesson for science policy: ‘Given the obvious benefits of deep learning to the scientific enterprise, it’s worth asking why it took OpenAI to push the field forward… and not, say, the National Science Foundation or one of America’s world-class research universities? To be sure, most of the intellectual breakthroughs behind our current AI summer originated in academic research…. A close cousin to transformer models was first published about in the early 1990s, along with a methodology for unsupervised pre-training. Yet as any graying AI researcher will tell you, neural networks were a backwater of the field until relatively recently…. Grant-funded science rewards novelty, while the core architectures behind recent progress in AI are anything but. Imagine a research proposal that said something like the following: “We are requesting one billion dollars to train a really big neural network. We don’t anticipate making any major theoretical advances in algorithm design, and we don’t yet know the exact details of our plan, much less if it will work. But if there’s even a small chance that scaling existing models up will reveal a path to AGI, we think it’s worth the shot.” Such a proposal would be dead on arrival, yet it is essentially what OpenAI led with…. Focused Research Organizations (FROs) are science and engineering programs addressed to “well-defined challenges that require scale and coordination but that are not immediately profitable.” The case for FROs in the context of federal funding for science and R&D was made succinctly by Sam Rodriques and Adam Marblestone…