Razib Khan's Unsupervised Learning

Katherine Brodsky hosts the Substack Random Minds and is author of the soon-to-be-published book No Apologies: How to Find and Free Your Voice in the Age of Outrage―Lessons for the Silenced Majority. The daughter of Ukrainian Jewish immigrants to Israel and then Canada, Brodsky has worked as a photographer, in public relations and as a publisher. A recent visiting fellow of the Danube Institute, she is freshly back in North America in the wake of Israel's Gaza invasion, following the Hamas attacks of 10/7.

Razib and Brodsky have a wide-ranging conversation about her travels in Europe, her return to North America, her understanding of Israel and what it's like to be a Jew in the world today. Despite Hungary being ruled by a right-wing government for the past decade, Brodsky observes that it is the European nation where she perceived the least public anti-Semitism, in large part due to government policy suppressing such sentiments. In contrast, she felt very uncomfortable and even afraid in London when she visited during the mass protests in support of the Gazans. Brodsky has also been shocked by the unanimity of the global Left in support of the Palestinians, and the dismissal by many feminist organizations of the reality of sex crimes committed by Hamas on 10/7. A secular Jew, Brodsky has now begun to wear visible signs of her Jewishness as a rebuke to the anti-Semitism she sees all around her.

Razib asks her about her opinions in regard to free speech and the battle between pro-Israel and pro-Palestinian factions in the media. Brodsky believes that the topic of her book, No Apologies, is even more relevant after the events of the last few months than it was when she initially wrote the book proposal. They discuss the lack of dialogue, debate and the polarization into strict information tribes. When it comes to information, they touch upon semantic arguments about indigeneity, who is a settler in the context of Israel-Palestine, and how it relates to genetics, history and archaeology.

Today, Razib cross-posts an episode of his other podcast. When not working on this Substack, Razib devotes his time to GenRAIT, a startup accelerating scientific discovery by providing infrastructure and tools to researchers. GenRAIT fosters science and discovery by making biological data accessible, usable, and minable.

Razib and his cofounders, Dr. Santanu Das and Taylor Capito, will talk about what they've built at the JPM Healthcare Conference in San Francisco, January 8th-11th, and showcase their products at the Plant and Animal Genomes Conference in San Diego, January 12th-17th.

After an overview of GenRAIT's business logic with Capito, Razib digs deeper with Dr. Das, a computer scientist by training, and discusses what machine learning is, drilling down into details like the difference between SVM and random forest models. Dr. Das surveys the multi-decade arc of "Artificial intelligence" (AI), predicts its inevitable utility in biology, and makes the broader case for productizing scientific tools to empower bioinformaticians, ML engineers and data scientists as an integral part of the scientific ecosystem.

On this episode of Unsupervised Learning Razib talks to Philippe Lemoine, a fellow at CSPI, a philosopher of science trained at Cornell. Lemoine often wades into controversial topics, like whether Chinese COVID data is trustworthy, but recently, he posted on Twitter that "Americans *genuinely* believe they have better food than France. They really believe it." Not only did this trigger a response by Tyler Cowen at Marginal Revolution, but the controversy broke out of social media into the international media.

For the first portion of the conversation, Razib and Lemoine reflect on the circus surrounding his tweet, and what he really means. Though Lemoine defends his tweet, he wants to emphasize that while Americans kept pointing to restaurants, he wanted to emphasize how superior French home cooking was over what Americans produce when they eat in (though he will still defend French restaurants as superior, he admits that if you value variation and diversity of cuisine as goods in and of themselves, then "American food" can be counted as superior). Additionally, Lemoine extolls the virtue of French meal-time customs and norms, with their leisurely pace. He asserts, likely plausibly, that Americans with their on-the-go philosophy of life truly don't enjoy food, they consume it. Razib pushes Lemoine on whether he would genuinely prefer to live in France despite the nation being 33% poorer than the US on a purchasing-power-parity basis, and he sticks to his guns.

Then they move to a topic that unambiguously throws France into a worse light than the US: immigration policy. Lemoine discusses the reality that French immigrants and immigrant-descended populations are not doing very well. Not integrating, committing crime and both economically and socially marginalized. He also claims that the French government actually does have a good breakdown of ethnic groups even though it cannot officially collect such data. Lemoine asserts that America's positive experience with immigration has little to do with America, and mostly to do with the source and character of immigrants: unlike Europe, the US imposes a strong selective sieve on education and skills. In contrast, European nations often simply receive immigrants from their former colonies or through the asylum process. Razib and Lemoine also discuss French fertility, which remains higher than that of the US.

On this episode of Unsupervised Learning Razib discusses war and diplomacy from 9/11 to 10/7 with Mark Safranski. Safranski is a long-time military affairs and foreign policy commentator who ran the popular weblog Zenpundit beginning in 2003. They survey how the 21st century, from the 9/11 attacks down to the Hamas atrocity against Israelis on 10/7, has seen a transformation of war and diplomacy by other means. From an age of flip phones as a luxury item in the early 2000's to ubiquitous smartphones in the 2020's, Safranksi and Razib explore how even still in 2023, Russia and Ukraine are engaging in the most significant set-piece battles since War War II, recapitulating some of the worst aspects of stagnant trench warfare more than a century on.

Safranski then explains the theories promoted by military scholars about how warfare has, and should, change in the technological era, particularly William Lind's ideas of fourth and fifth-generation warfare. He also argues that drone technology in various forms has been around since World War I, though it was perfected in the past twenty years. Considering America's technological advancements over the 21st century, Safranski alarmingly explains that the gap between us and our nearest competitor, China, is rapidly closing (with the US losing every government-sponsored war game against China in the years around 2020). He believes that the unipolar moment of the late 1990's is truly ending, and elaborates on the decay that has overtaken some branches of the US military, in particular, what had been the world's dominant blue-water navy, which is literally rusting away.

They conclude the conversation by reflecting on the changing role of the blogosphere in influencing military and foreign policy thinkers, and how the decade after 9/11 saw a fertile cross-pollination between online discourse and the brain trusts of the military-industrial complex. Finally, Razib asks Safranski if the neoconservative movement is making a comeback in the wake of Russia's Ukraine invasion, and what has surprised him about American reactions to 10/7.

On this episode of Unsupervised Learning, Razib discusses personality with Brent Roberts, professor of psychology at the University of Illinois. Roberts explains what personality actually is as a psychological construct, and how it differs from personality traits, like extraversion. Razib and Roberts also address the Big Five Personality system, and how it relates to the Myers-Briggs framework. Roberts elucidates what the Big Five's extraversion, openness, conscientiousness, neuroticism and disagreeability actually mean, and how they correspond to Myers-Briggs dichotomies of extraversion/introversion, sensing/intuitive, thinking/feeling and judging/perceiving. Though Myers-Briggs may have some shortcomings, for example collapsing continuities in discrete categories, Roberts maintains that it still retains some utility, and is not as unscientific as many researchers assert.

Razib then asks about personality's relationship to behavior genetics. What is the heritability of personality's subcomponents? How do they interrelate? And what might the evolutionary context of personality variation be? Roberts also addresses the idea that personality traits can change over one's lifetime, and are unstable from test to test. Razib also wonders how different traits, like agreeability, correlate with life outcomes in income and happiness. Finally, they discuss the nature of psychopaths and sociopaths, and how they relate to the Big Five categories.

On this episode of Unsupervised Learning Razib talks to Carl Zha. Zha is a Sichuan-born China-commentator who had a long-term professional sojourn in southern California, before settling in Bali, Indonesia. He hosts the Silk and Steel podcast, which covers China, the Silk Road, and more general history, culture and geopolitics. Zha and Razib have known each other since the 2010's, and often circle back to discussions of China, its history, politics and culture. The course of their conversations has spanned both the close of the "Chimerica" period of trade and political relations, and the more adversarial status that obtains in both the US and in the People's Republic of China under Xi Jinping.

But first, Razib and Zha discuss what it's like to live as an ethnic Han Chinese in Indonesia, albeit one who resides in Hindu-majority Bali, where Zha settled after marrying a local woman and becoming a father. Though Indonesia has an economically and politically powerful Chinese minority, it was also the scene of ethnic riots in the 1990's and a genocide of Han Chinese in the 1960's. Until recently, the state did not recognize Confucianism as a religion and discouraged Chinese names and Chinese-language schools. Nevertheless, Zha presents a relatively positive picture of relations on the island of Bali, where the Hindu population seems to have had an easier time integrating Han in a more syncretistic culture than in the Muslim-majority islands.

Then they discuss the pivot in US-China relations in the last half a decade, and the possibilities presented by great power rivalry. Razib and Zha address the thorny reality that though China and the US are now embarking on more explicitly adversarial geopolitics, their economic ties remain strong, with Chinese supply chains essential for American firms like Apple and the US consumer demand essential to propping up China's vast export sector. Zha also offers a defense of Xi Jinping's rule and the prospects for China as it turns inward from the world, focusing on its domestic market and shoring up its geopolitical positions. The discussion turns to the range of likely outcomes in a world where the 21st century is both the American and Chinese century, and the two great powers remain both economically and geopolitically entangled through trade and numerous bilateral relationships with other nations.

Today, Razib interviews Nikolai Yakovenko, already a three-time guest on his podcasts (A Twitter engineer on machine learning and his former company's prospects, GPT-3 and the rise of the thinking machines and AI and Biology). An artificial intelligence researcher based in Miami who has worked at Google and Nvidia, Yakovenko is the founder of DeepNews where he currently works.

Razib and Yakovenko talk about the economic, technological and socio-political implications of the leadership turmoil at OpenAI, the $86 billion dollar company that has supercharged the field of artificial intelligence with their product, ChatGPT. Yakovenko digs deep into the nuts and bolts of how artificial intelligence works today, from transformers to GPUs and different compute needs of training vs. inference. They also discuss the importance of shipping products at a tech company, in contrast to simply publishing papers as the measure of productivity in academic basic science research for example, and how on this count OpenAI succeeded where Google has not. Razib asks Yakovenko if OpenAI might now fall behind Google, whose corporate risk-aversion had squandered it an early lead in technology like transformers. Now that OpenAI is seized by such organizational chaos as to stop development, Google may have time to catch up. Yakovenko also talks about the likelihood of artificial intelligence becoming a corporate oligopoly due to the field's colossal compute needs. Finally, Razib and Yakovenko address the cleavages that arose at OpenAI due to the board's adherence to effective altruism, while leadership and employees instead charted a shift toward effective accelerationism.

On this episode of Unsupervised Learning, Razib talks with Curtis Yarvin. The host of the Grey Mirror Substack, Yarvin is the former Mencius Moldbug, a pseudonym under which he wrote extensively on culture, politics and history. Yarvin's social and political views have been profiled widely, including by Vanity Fair and Vox. The intellectual father of neo-reactionary thought, Yarvin is also trained as a computer scientist, and in 2010, he released the first version of Urbit, a decentralized personal server platform, which has spawned an entire community and conferences like Urbit Assembly.

Yarvin's interests extend beyond technology and politics. He is deeply invested in high culture and believes in the importance of the humanities to our civilization. Razib and Yarvin spend most of this episode on the role of poetry in our broader culture, why it is relevant, why it matters, and the works that Yarvin most values. Though his educational background is as a technologist, Yarvin believes that poetry is an essential ingredient in what makes us human. The question of humanity, along with some references to the Dune universe, moves the conversation to Yarvin's reflections on the rise of A.I. via large language models, and whether it poses an existential threat to the human race (he does not believe it does). Yarvin also offers his opinions on Eliezer Yudkowsky's trajectory of thought; Yudkowsky has in the last decade become the leader of the "A.I. doomer" faction within the futurist community. Like many observers with a strong computational background, Yarvin does not believe A.I. will lead to the abolition of man, though it may open up new possibilities, extending what humans are capable of in terms of innovation and productivity through cybernetics. Like all technology, its ultimate utility will be contingent on our decisions as humans.

On this episode of Unsupervised Learning, Razib talks to Michael Muthukrishna about his new book, A Theory of Everyone: The New Science of Who We Are, How We Got Here, and Where We're Going. Muthukrishna is Associate Professor of Economic Psychology at the London School of Economics, an affiliate of the Developmental Economics Group at STICERD and Data Science Institute, Azrieli Global Scholar at the Canadian Institute for Advanced Research (CIFAR), Technical Director of The Database of Religious History, a fellow at the Charter Cities Institute and board member of the One Pencil Project. Of Sri Lankan extraction, he trained as an engineer in Australia, but later became interested in anthropological and cultural questions. He studied for his Ph.D. under Joe Henrich in Canada. Like his mentor, Muthukrishna cross-applies toolkits from evolutionary biology and population genetics to questions of variation and change in human cultures.

A Theory of Everyone is an ambitious book with arguably galactic ambitions. The chapters jump from topics like the Cambrian Explosion to the ever-increasing amount of energy needed to get at the fossil fuels that power our civilization. But to start off, Razib asks Muthukrishna about his background as a "third culture kid" and how that might have influenced his anthropological interests. Muthukrishna observed firsthand social and political chaos in Papua New Guinea, while his family's background in Sri Lanka illustrated for him the salience of ethnic tensions, even when differences might seem minimal to outsiders. Then Razib talks about A Theory of Everyone's fixation on energy and its role in powering organic life, about our technology-driven civilization and about our potential interplanetary future. Here, Muthukrishina thinks like an engineer, albeit with a broad historical and evolutionary perspective. He and Razib also discuss the problems of "degrowth economics" and why it is a dead-end for a dynamic civilization's flourishing. Razib also probes Muthukrishna for his views on IQ, its utility as a psychological measure, the variation between individuals and groups, and how those might relate to cultural evolutionary frameworks for considering cognitive aptitudes. The conversation concludes with a consideration of future possibilities as we hurtle past our current energy constraints as a civilization (Muthukrishna is bullish on nuclear), and the role of decentralized political experimentation in improving our social technology.

On this episode of Unsupervised Learning Razib talks to Peter Nimitz about what he memorably calls the crisis of the 23rd century. Most people know of the fall of Rome, and the subsequent European Dark Ages. And because of scholars like Eric Cline, today growing numbers are aware of the civilizational collapse at the end of the Bronze Age, when an incipient global civilization enfolding everything from the shores of the eastern Mediterranean to Mesopotamia was torn apart by climate change and invasion. But before Rome, before the Hittites, the third millennium BC saw a climatic shock that seems to have abruptly transformed all Eurasia and Africa. There have long been glimmers of this upheaval; historical texts record chaos in Sumer and Akkad, while in Egypt the Old Kingdom fell. But today the toolkit of archaeology can illuminate far more, and is making clear that a massive climatic shift toppled fragile empires and transformed cultures. Some areas of Eurasia, like China, seem to have experienced massive drought. Others, like Siberia, became even colder, with regions becoming ever more inhospitable to human occupation.

And now Nimitz has reviewed what we know region by region in a magisterial post, Crisis of the 23rd Century: Upheavals from Spain to the Yangtze. Razib presses him about the contrast between the three peninsulas of Southern Europe: the Iberian, Italian and Balkan, and how they each experienced the intrusion of Indo-Europeans during this period (informed by ancient DNA findings). They also discuss the potential divide between Corded Ware and non-Corded Ware Indo-European migrations, and how that shook out across Eurasia. Then they touch upon the civilizational hearths of the Near East, and the wholesale transformation that occurred at this moment culturally and demographically in the Horn of Africa. Finally they sweep eastward, into Siberia, Central Asia and finally China, where new cultures arose and old civilizations collapsed in the centuries after the crisis of the 23rd century.