Michelle Ma
0 followers · 849 views
on the atlas — 38
- The rare people who are solid - by Sasha Chapin5 savers
- Context Widows - by Kevin Baker - Artificial Bureaucracy2 savers
- The Biotech Delusion: Things No One Wants to Admit1 savers
- I'm skeptical that powerful AI will solve major human problems1 savers
- John von Neumann Shot Lightning From His Arse1 savers
- How to Incubate a Dream | Dream Studies Portal1 savers
- On the limits of idealized values - Joe Carlsmith2 savers
- On clinging - Joe Carlsmith8 savers
- The MTG Color Wheel (& Humanity) - by Duncan Sabien1 savers
- On attunement - Joe Carlsmith3 savers
- Gmail - Miércoles de ceniza4 savers
- Inside the Lobbying Frenzy Over California's AI Companion Bills | TechPolicy.Press1 savers
- Is China Serious About AI Safety? | AI Frontiers1 savers
- Stewardship: CEA’s 2025-26 strategy to reach and raise EA’s ceiling — EA Forum1 savers
- What is intelligence? - by vasto and Asterisk Magazine1 savers
- "Be It Enacted" - by Dean W. Ball - Hyperdimensional1 savers
- Full text of "Are Inventions Inevitable? A Note on Social Evolution"1 savers
- What Technology Wants Summary by Kevin Kelly | Jon Santiago1 savers
- Nanotechnology’s spring - Works in Progress3 savers
- Credit rating agency - Wikipedia1 savers
- How Should AI Liability Work? (Part I) - by Dean W. Ball3 savers
- The Six Foundational Books to Read on China’s Political Economy2 savers
- You can just things - by Hyde - BreatheLess1 savers
- Buteyko Method 101 - by Hyde - BreatheLess1 savers
- Hypervigilance - by Hyde - BreatheLess1 savers
- Why Warm Countries Are Poorer - by Tomas Pueyo4 savers
- Guide world5 savers
- Reflections from Kunming, an unglobalized part of the world3 savers
- Notes on Guinea – Matt Lakeman1 savers
- Notes on Mauritania – Matt Lakeman1 savers
- Notes on Tajikistan – Matt Lakeman4 savers
- Arcadia (play) - Wikipedia2 savers
- Dario Amodei — Machines of Loving Grace34 savers
- Laws of Tech: Commoditize Your Complement · Gwern.net26 savers
- LLM Daydreaming · Gwern.net12 savers
- So Much Rejection6 savers
- How to instantly be better at things - by Cate Hall6 savers
- What Are We Trying to Do? - by yung macro 宏观年少传奇3 savers
highlights — 139
What stops congruence from occurring are layers of denial that are unpleasant to pass through. Although congruence is a source of endless happiness, the path there can be devastating. To paraphrase a cliche, you may have to finally give up on experiencing a better past.
The rare people who are solid - by Sasha ChapinBut that would still require policy change. And my whole point here is that just because you can identify a beneficial policy change, that doesn’t mean it will happen.
I'm skeptical that powerful AI will solve major human problemsI think you find that there are huge barriers to accomplishing anything outside of the narrow space of media/software intangibles.
I'm skeptical that powerful AI will solve major human problemsGiving such outsiders a shot actually been tried, with the various ‘superforecaster’ experiments in AI predictions, which I’ve analyzed extensively. For various reasons, including broken incentives, you end up with both timelines and risk levels that I think of as Obvious Nonsense, and we’ve actually spent a decent amount of time grappling with this failure.
Gmail - Miércoles de cenizaAs in, I think it’s pretty patently absurd to expect it almost certain that, if we construct new minds generally more capable than ourselves, that this turns out well for the humans. Also, why would they want to work there, and even if they do, how are they going to do the technical work?
Gmail - Miércoles de cenizaA particular stumbling block for US-China exchange came in March when BAAI was placed on the US Entity List, a designation generally justified by an organization’s ties to foreign military activity, and which prohibits organizations under US jurisdiction from trading with or transferring technology to the listed entity.
Is China Serious About AI Safety? | AI FrontiersChina’s international approach increasingly centers on the global diffusion of AI
Is China Serious About AI Safety? | AI FrontiersChina’s “Global AI Governance Action Plan,” which offers a competing vision to the Trump administration’s AI Action Plan, stresses the importance of international collaboration and broadly distributed capacity-building.
Is China Serious About AI Safety? | AI FrontiersCnAISDA is designed primarily as China's voice in global AI governance discussions
Is China Serious About AI Safety? | AI FrontiersThe organization currently lacks dedicated staff, a budget of its own, or even a Chinese-language website, raising questions about its operational capacity. The nature of its relationship with the Chinese government remains ambiguous, though an official from the powerful National Development and Reform Commission has publicly confirmed CnAISDA’s government backing.
Is China Serious About AI Safety? | AI Frontiersalso incorporates other leading technical organizations in AI
Is China Serious About AI Safety? | AI FrontiersTsinghua University serves as the organizational hub:
Is China Serious About AI Safety? | AI Frontiersfollows the approach taken by the Canadian AISI and France’s INESIA
Is China Serious About AI Safety? | AI FrontiersCnAISDA is a coalition of existing institutions.
Is China Serious About AI Safety? | AI FrontiersAndrew Yao, China's sole Turing Award winner
Is China Serious About AI Safety? | AI FrontiersCnAISDA’s leadership includes some of the preeminent figures in Chinese AI policy
Is China Serious About AI Safety? | AI FrontiersShanghai AI Lab, in partnership with Concordia AI, published a Frontier AI Risk Management Framework laying out a detailed process for monitoring and managing risks from frontier AI systems
Is China Serious About AI Safety? | AI FrontiersHowever, some key stakeholders, notably including DeepSeek, were conspicuously absent from the second round of commitments. A key indicator for safety concerns in China will be whether Chinese developers comply with their commitments
Is China Serious About AI Safety? | AI FrontiersIn December 2024, China's Artificial Intelligence Industry Alliance, a part of CAICT, released safety commitments signed by 17 major Chinese AI companies, including DeepSeek, Alibaba, and Tencent. Notably, these commitments closely mirror the Frontier AI Safety Commitments made by AI developers around the world at the Seoul AI Summit
Is China Serious About AI Safety? | AI Frontiersnewest elements of that ecosystem: the China AI Safety and Development Association (CnAISDA). Launched in February 2025 on the sidelines of the Paris AI Action Summit, CnAISDA places China among a small number of jurisdictions with dedicated AI safety institutes, or AISIs
Is China Serious About AI Safety? | AI FrontiersWhen I say “hard”, I mean “computationally hard”. The space of plausible hypotheses is so large that it takes an extremely long time to search through. Moreover, experiments often can’t reliably distinguish between many even qualitatively different hypotheses. (See also: identifiability.) Consider models of the solar system. It wasn’t just that people were stupid and superstitious, and simply couldn’t accept that the Earth revolves around the Sun—although that was certainly part of the issue in some times and places. Rather, a model where the planets and stars orbit the Earth, and a model wher…
What is intelligence? - by vasto and Asterisk MagazineGiven this, I’d modify Tegmark’s definition slightly so that it reads intelligence = ability to accomplish complex goals under constraints The “under constraints” part is important, since it’s what makes oracles and rulebooks less interesting, and people and neural networks more interesting.
What is intelligence? - by vasto and Asterisk Magazineit may be even more memory-efficient to memorize rules for generating rules. The process of generating a rule is called “learning” it.
What is intelligence? - by vasto and Asterisk MagazineThese kinds of rules are what we mean by computations: the forms of intelligence we’re familiar with don’t simply memorize a large number of input-output pairs, but instead transform an input into the corresponding output by following some sequence of steps. The sequence of steps, or algorithm, is itself constrained in various interesting ways.
What is intelligence? - by vasto and Asterisk MagazineThese limits incentivize compression, for example by generalizing from one input-output pair to similar ones, or by memorizing a rule that summarizes the often exponentially many input-output pairs rather than memorizing the pairs themselves.
What is intelligence? - by vasto and Asterisk MagazineConstraints are the reason the book we just imagined can’t exist, and (at least in my own view) are the reason that studying intelligence is interesting at all. In a magical world with no constraints (whatever that precisely means), there’s nothing that prevents Live-Your-Best-Life books and genies and oracles from existing. This insight is deeper than it first appears, because constraints yield a lot of the things we typically associate with intelligence.
What is intelligence? - by vasto and Asterisk MagazineIntelligence in this sense is often called “narrow” (or “weak” if you want to be more derogatory), since it concerns a specific goal or set of goals.
What is intelligence? - by vasto and Asterisk MagazineMax Tegmark agrees, and in his book Life 3.0 essentially defines intelligence in terms of capabilities (pg. 50): intelligence = ability to accomplish complex goals
What is intelligence? - by vasto and Asterisk MagazineI think “intelligence” is best understood as a matter of degree rather than as something that is present or absent.
What is intelligence? - by vasto and Asterisk MagazineMost of the signs of intelligence we’re familiar with can be quantified in terms of continuous performance measures.
What is intelligence? - by vasto and Asterisk MagazineMany things we might not naively consider as intelligent, like our ability to stand on one leg, or recognize a friend’s face, have been found to be quite computationally difficult to achieve—and hence, intelligent in at least that sense.
What is intelligence? - by vasto and Asterisk MagazineIt does not regulate the development of the technology itself and instead requires developers to tell the public how they are building the technology.
"Be It Enacted" - by Dean W. Ball - HyperdimensionalI knew this would be a major challenge when I was in government, and since leaving my doubts about the political prospects of federal preemption have only grown.
"Be It Enacted" - by Dean W. Ball - HyperdimensionalIn 1975, SEC rules began explicitly referencing credit ratings.[23] For example, the commission changed its minimum capital requirements for broker-dealers, allowing smaller reserves for higher-rated bonds; the rating would be done by "nationally recognized statistical ratings organizations" (NRSROs). This referred to the "Big Three",[24] but in time ten agencies (later six, due to consolidation) were identified by the SEC as NRSROs.[14][25]
Credit rating agency - WikipediaIn 1936, regulation was introduced to prohibit banks from investing in bonds determined by "recognized rating manuals" (the forerunners of credit rating agencies) to be "speculative investment securities" ("junk bonds", in modern terminology). US banks were permitted to hold only "investment grade" bonds, and it was the ratings of Fitch, Moody's, Poor's, and Standard that legally determined which bonds were which. State insurance regulators approved similar requirements in the following decades.[14]
Credit rating agency - WikipediaCredit rating agencies originated in the United States in the early 1900s, when ratings began to be applied to securities, specifically those related to the railroad bond market
Credit rating agency - WikipediaYet we also have reason to believe that tort liability led to unintended consequences. As workplace harm lawsuits mounted, discrimination in employment against pregnant women, the disabled, and other protected classes became common, since these groups were, for various reasons, at higher risk of generating tort liability exposure (in some occupations, pregnant women were banned from employment until this practice was made unlawful). Vaccines were made almost impossible to release without federal carve outs from liability (as was required for the Covid vaccines). Many vaccine companies went out…
How Should AI Liability Work? (Part I) - by Dean W. BallWe know for a fact that anxious people perform much worse in anti-saccadic tests. And we also know that CO2 and hyperventilation are tell-tale signs of anxiety. These connections point to how breathing plays a role in generating excessive response to stress. Someone with calm, slow breathing and higher CO2 tolerance2 will fare better in challenging situations but more importantly, they’ll be able to return to the baseline of security much faster. Someone predisposed to hyperventilation has a lower threshold for getting overstimulated, nervous, or exhibiting other neurotic traits — it takes les…
Hypervigilance - by Hyde - BreatheLessThen again, after some thought, one considers that car right-of-way really is the efficient Coasean solution here. Given the number of pedestrians, it’s correct to place the cost on those who can stop most efficiently. Otherwise there would be no way for cars, buses, or bikes to get around at all.
Reflections from Kunming, an unglobalized part of the worldMost of all we should avoid this tendency identified by Thiel: “People are increasingly pessimistic about the existence of new and interesting things. Can we go to the moon? We’ve done that already. Mars? Impossible, many people say. What about chemistry?… The periodic table seems pretty set. It may be impossible to discover anything new there. The frontier is closed. There is nothing left to discover.”
Reflections from Kunming, an unglobalized part of the worldThiel has used an entire lecture to remind us the importance of belief in secrets: “The people who actually solve hard problems are people who believe in secrets. If you believe something is hard, you might still think you can do it. You’ll try things, and maybe you’ll succeed. But if you think something is impossible, you won’t even try.”
Reflections from Kunming, an unglobalized part of the worldIf you don’t want people to be discouraged, maybe it’s better they don’t know of all the development already out there. The tradeoff is sometimes you waste the efforts of people who re-invent various wheels. But from a learning point of view, that may not be so negative. I submit that the process of manually working through solved problems is an underrated learning experience. Sometimes I re-do manual calculations of math problems usually trivial to solve; I used to make a habit out of re-typing various magazine articles (usually from the New Yorker) because it made me hyperaware of sentence c…
Reflections from Kunming, an unglobalized part of the worldAs part of his idea of “innovation starvation,” Neal Stephenson has written on how it’s become much easier to be discouraged from trying various things. If one comes up with a novel idea, it’s very common to search on the Internet to see if it’s been tried before. And usually it looks like it sort of has. That’s typically discouraging, and one drops the idea of developing something novel.
Reflections from Kunming, an unglobalized part of the worldHere’s Stephenson: “What if that person in the corner hadn’t been able to do a Google search? It might have required weeks of library research to uncover evidence that the idea wasn’t entirely new — and after a long and toilsome slog through many books, tracking down many references, some relevant, some not. When the precedent was finally unearthed, it might not have seemed like such a direct precedent after all. There might be reasons why it would be worth taking a second crack at the idea, perhaps hybridizing it with innovations from other fields.”
Reflections from Kunming, an unglobalized part of the worldChristianity is primarily based around the teachings of a pacifist hippy who famously preached a spiritual separation of church and state with, “Render to Caesar the things that are Caesar’s, and to God the things that are God’s.” In contrast, Islam is based on the teachings of a head-of-state (ie. warlord) whose words and actions were said to be the infallible projections of god. Hence, Islam as a doctrinal belief system has a lot more to say about the workings of government than Christianity or the vast majority of religions. This has resulted in an unusually intense intertwining of state an…
Notes on Tajikistan – Matt LakemanDushanbe hosts what was once the largest flagpole in the world, constructed with a mere $3.5 million. It is 165 meters high, which I know not through Wikipedia, but because four separate Tajiks proudly quoted the figure to me without prompting.
Notes on Tajikistan – Matt Lakeman(Note – like Uzbekistan and Kyrgyzstan, Tajikistan was largely relegated to cotton production by the USSR despite cotton being extremely water-intensive and the region having very little water [thus leading to the death of the Aral Sea]. Cotton is still a fairly significant portion of the Tajik economy, but its proportional value has been waning, and in 2023, the cotton industry was a net-loss for the country.)
Notes on Tajikistan – Matt LakemanBy 2022, China held $2 billion in Tajikistani debt, more than half its total debt at the time. What has Tajikistan been doing with all this borrowed money? In the early 2010s, the Tajik government spent more than $210 million (more than 1/10th of its annual budget) on developing Dushanbe, but the vast majority of this funding went toward big government palaces and monuments (more on that later). I can’t find good data on recent agreements between Tajikistan and China, but my speculation is that there has been a dramatic ramp-up in funding for the former over the last few years. According to al…
Notes on Tajikistan – Matt LakemanIn all these instances, America has pursued federal legal standards. In all these instances, America has eliminated or severely limited the exposure of the foundational technology providers to tort liability for downstream misuse of their products.
Where We Are Headed (Part II) - by Dean W. BallThis jumped out instantly when I looked at the charts: Your prior and evidence can't possibly both be correct at the same time. Everywhere the prior has non-negligible density has negligible likelihood. Everywhere that has substantial likelihood has negligible prior density. If you try multiplying the two together to get a compromise probability estimate instead of saying "I notice that I am confused", I would hold this up as a pretty strong example of the real sin that I think this post should be arguing against, namely that of trying to use math too blindly without sanity-checking its meanin…
Why We Can't Take Expected Value Estimates Literally (Even When They're Unbiased) — LessWrong