Sam B
13 followers · 8 following · 361 views
on the atlas — 44
- [2103.00065] Gradient Descent on Neural Networks Typically Occurs at the Edge of Stability1 savers
- [2510.19315] Transformers are Inherently Succinct1 savers
- [2607.03497] Isolated Hypersurface Singularities May Be Stably Degenerate1 savers
- Early-stage empirical work on “spillway motivations” — LessWrong2 savers
- Advice for Making Life More Whimsical & Joyful1 savers
- Scaling Laws, Carefully | Lil'Log15 savers
- A reading list for generalists — LessWrong18 savers
- The age of average — Alex Murrell12 savers
- Every Frame Perfect @ tonsky.me1 savers
- Competing with sampling — Alignment Research Center5 savers
- AI-Enabled Coups: How a Small Group Could Use AI to Seize Power11 savers
- My two-part desk setup1 savers
- Making Deep Learning Go Faster29 savers
- Opinion | Silicon Valley Is Bracing for a Permanent Underclass - The New York Times4 savers
- Maybe I was too harsh on deep learning theory (three days ago) — LessWrong5 savers
- The Making of a Techno-Nationalist Elite - American Affairs Journal5 savers
- Single Token Geometry: DeepSeek V4 and Manifold Tearing2 savers
- the jackpot age — LessWrong1 savers
- Sam Altman May Control Our Future—Can He Be Trusted? | The New Yorker10 savers
- Introducing nanocode: The best Claude Code that $200 can buy. · salmanmohammadi/nanocode · Discussion #11 savers
- Julia10 savers
- The Outliers - by Joseph Heath - In Due Course1 savers
- [2602.07852] Emergent Misalignment is Easy, Narrow Misalignment is Hard1 savers
- [2603.02202] Frontier Models Can Take Actions at Low Probabilities4 savers
- Growth and Form in a Toy Model of Superposition — LessWrong2 savers
- [2602.22631] TorchLean: Formalizing Neural Networks in Lean1 savers
- [2308.12108] The Local Learning Coefficient: A Singularity-Aware Complexity Measure4 savers
- Nobody ever got fired for using a struct1 savers
- Timaeus | Learn about SLT1 savers
- SLT for AI Safety — LessWrong1 savers
- Singular Learning Theory Seminar2 savers
- Investigating the learning coefficient of modular addition: hackathon project — LessWrong2 savers
- Rust GPU1 savers
- Something is Going on with Colorectal Cancer1 savers
- Dispatches - VectorWare1 savers
- Engram/Engram_paper.pdf at main · deepseek-ai/Engram1 savers
- The housing theory of everything - Works in Progress34 savers
- Modular Manifolds - Thinking Machines Lab16 savers
- First, Make Me Care, by Gwern · Gwern.net11 savers
- When Models Manipulate Manifolds: The Geometry of a Counting Task9 savers
- Making Software: How is data stored?7 savers
- PyTorch internals : ezyang’s blog7 savers
- [2412.06769] Training Large Language Models to Reason in a Continuous Latent Space6 savers
- Against the dark forest3 savers
highlights — 31
In this sense, A.I.’s broad capabilities foster a rare class solidarity between white-collar and blue-collar workers. When 20-something software engineers in San Francisco talk about escaping the permanent underclass, I hear them projecting concerns about their own precarity: What happens if the invisible hand of the market decides that my skills are no longer valuable? Who will catch me if I fall? For once, a rarefied class of employees — those used to being the automaters, not the automated — is reckoning with their potential obsolescence.
Opinion | Silicon Valley Is Bracing for a Permanent Underclass - The New York TimesIf A.I. companies and American voters are waiting for policymakers to act, many still seem paralyzed by the data (or lack thereof). Nobody can predict how far A.I. capabilities will progress or how fast A.I. will spread. Economists also disagree on whether wage inequality will rise or fall, whether consumer demand is elastic or capped and whether economic growth will be linear or exponential. As a result, many are hesitant about making aggressive forecasts, even in scenarios in which advanced A.I. rapidly surpasses human ability.
Opinion | Silicon Valley Is Bracing for a Permanent Underclass - The New York TimesUnder a disco ball, with La Croix cans in hand, tech workers perched on benches and beanbags while Mr. Shor presented a slew of public opinion polls spotlighting Americans’ economic fears about A.I. One slide showed that 79 percent of voters are worried about the “government not having a plan to protect workers,” and 72 percent are concerned that A.I. “drives down wages for people like you.”
Opinion | Silicon Valley Is Bracing for a Permanent Underclass - The New York TimesWhat I found was a well of worry, good ideas and limited commitments from corporate actors whose core business model relies on the very disruption they are warning about.
Opinion | Silicon Valley Is Bracing for a Permanent Underclass - The New York Times(In general, tech industry sources expressed more extreme concern about the labor market impacts of A.I. in private conversation — but suddenly became optimists once I turned on the mic.)
Opinion | Silicon Valley Is Bracing for a Permanent Underclass - The New York TimesAmerica’s first techno-nationalist elite did have such a project. Many of them died fighting for it. The industrial civilization they built would have been impossible without their ironclad commitment to America’s national greatness. Judged by that standard, Karp and Zamiska’s arguments are intolerably thin. “Those who say nothing wrong,” Karp and Zamiska warn, “often say nothing much at all.”88 The Technological Republic says nothing wrong and nothing much at all.
The Making of a Techno-Nationalist Elite - American Affairs JournalWhat principles should govern the relationship between artificial intelligence and the American polity? Does transhumanism violate or embody the “shared purpose and identity” Karp and Zamiska believe we must forge? If what we need is a “larger project for which to fight,”87 then what precisely should that project be?
The Making of a Techno-Nationalist Elite - American Affairs JournalThey demand a pulpit for America’s technologists but never summon the courage to state what gospel they should preach.
The Making of a Techno-Nationalist Elite - American Affairs JournalKarp and Zamiska recognize that Silicon Valley’s engineering elite lack the cultural confidence to defend any way of life. In a revealing passage, they lament that America’s technologists shy away from “the vital yet messy questions of what constitutes a good life, which collective endeavors society should pursue, and what a shared and national identity can make possible.”79 This fundamentally misdiagnoses the problem, however. Silicon Valley’s failing is not that its leaders refuse to ask such questions: it is that they refuse to answer them. Here Karp and Zamiska cannot escape the disease th…
The Making of a Techno-Nationalist Elite - American Affairs JournalThe Technological Republic suggests that the federal government could profit from Silicon Valley’s organizational ingenuity, but it does not suggest how elected officials or federal bureaucrats might gain that expertise firsthand. It argues that technologists must identify with the American nation-state, but it never explains how this culturally progressive, immigrant-heavy industry might actually do so.
The Making of a Techno-Nationalist Elite - American Affairs JournalAny governing class requires three things: a political coalition to which it owes allegiance and over which it exercises influence; an economic base that provides this class with wealth and unites its members around shared material interests; and finally, a set of institutions, rituals, and social customs that give this class a culture distinct from the country at large. Absent the first two, a leadership class lacks the power to lead; absent the latter two, it lacks the ability to act as a class. The Eastern Establishment’s seventy-year dominance rested on its possession of all three.
The Making of a Techno-Nationalist Elite - American Affairs Journalthey do not investigate the religious, social, political, or economic milieu that created these technologists.
The Making of a Techno-Nationalist Elite - American Affairs JournalThese observations may be accurate, but the goal of The Technological Republic is to inspire American technologists to become American techno-nationalists. Regrettably, Karp and Zamiska offer no roadmap for accomplishing this.
The Making of a Techno-Nationalist Elite - American Affairs Journalthe flattening of global diversity to fit the norms and interests of any given American techno-culture—corporate or otherwise—is both a baldly colonial aspiration and one we should scorn for the same reason that we leave the idea of effective, monolithic, planetary-scale government—benevolent or otherwise—to underbaked science fiction.
Against the dark forestGlobal mega-platforms under capitalism are structurally incapable of handling the business of civilization: of governance, of providing genuinely public infrastructure, of making knife-edge decisions about the balance of liberties and securities.
Against the dark forestFew, if any, of this moment’s apparently unstoppable tech platforms will survive for long. The people on them will eventually leave—when they’re forced to do so by the continuous degradation of their experience, or because they’re forced to do so because their governments put the hammer down,
Against the dark forestThis was always true, but maybe it's clearer now, as we watch the recreational troll armies and mega-platform leaders and openly supremacist policymakers and the next US presidential administration glop together like a bad special effect. But I would argue that socio-technical systems that invite and facilitate proto-fascist programs of universal surveillance, control, and dehumanization aren't safe in any hands.
Against the dark forestit seems questionable for technologists to cede the territory of the public internet to their fellow-but-worse technologists and the predatory forces they assemble and arm.
Against the dark forestThis all stops being an individual problem and becomes a collective one when bad products of the social internet get worse, as when platform turmoil and manipulation helps remodel the offline world in the image of the most grotesque parts of the online one.
Against the dark forestit positions the broad landscape of connection as something that “we” can simply do without—and without which we will indeed feel better and be more productive.
Against the dark forestInstead we are rapidly entering a time of instability and uncertainty, driven by the appearance of success (not success itself) on TV.
The Great Entertainment - by kyla scanlonThis is why “Make America Great Again” is so resonant. It promises a return to a moment that never existed for most people and cannot be recreated through performance alone. Nostalgia works as content but it certainly fails as governance.
The Great Entertainment - by kyla scanlonWhat comes after the Great Entertainment won’t be a return to the Great Moderation. But it might be something better: where material reality matters more than the TV scoreboard and where governance is measured by what it builds rather than how it plays. A return to the real.
The Great Entertainment - by kyla scanlonThe cameras keep rolling. Trump keeps performing. But the infrastructure that made the entire show possible - the dollar as reserve currency, the bond market, the alliances - appears to be already changing channels. Reagan proved you could use TV aesthetics in governance. Trump is proving you cannot replace governance with TV.
The Great Entertainment - by kyla scanlonThe President must be allowed to stir up drama5, and no one is allowed to fight back (or if they do, it should be in the name of Great TV). The Administration can go on TV and assure everyone that all of this is happening because of the Globalists, a made-up foil to advance the narrative, or because of the immigrants, or because of the UN or because of NATO - they can say anything and everything and they will. But the world is not a reality TV show. People’s lives cannot be governed by what makes ‘good TV’. It must be governed by trust. It must be governed by realness.
The Great Entertainment - by kyla scanlonChina now generates more electricity than the US and the EU combined, with 40 new nuclear power reactors under construction, compared to… zero in the United States. Energy is what matters, the real economy.
The Great Entertainment - by kyla scanlonIt fundamentally 100% does not matter who has the best AI girlfriend/boyfriend if they do not have a grid to support it.
The Great Entertainment - by kyla scanlonThe four key points to take away are:
DSLT 2. Why Neural Networks obey Occam's Razor — LessWrongnegative log likelihood (NLL), Ln(w),
DSLT 1. The RLCT Measures the Effective Dimension of Neural Networks — LessWrongWith this in mind, SLT provides a framework for understanding phases and phase transitions in neural networks. It has been mooted that understanding phase transitions in deep learning may be a key part of mechanistic interpretability, for example in Induction Heads, Toy Models of Superposition, and Progress Measures for Grokking via Mechanistic Interpretability, which relate phase transitions to the formation of circuits.
Distilling Singular Learning Theory - LessWrongif every single pundit or political figure who has ever referred to trans people as political liabilities—from Hillary Clinton to Bill Maher—was to asked a simple, binary question: are they too stupid to do basic due diligence on the issues they feel compelled to speak about, or are they concern-trolling bigots? It’s either one or the other. They are either incapable of doing the basic research that shows that Republicans are not succeeding on these attacks, which means these figures should never be seen as credible again. The alternative is that they know that Republicans are failing, and are…
The Modern Electoral History of Transphobia