Razib Khan's Unsupervised Learning

Today Razib talks to Theo Jaffee, co-host of Monitoring the Situation, a show backed by Andreessen Horowitz that covers the rise of the “intelligence explosion” and AI. Jaffee is writer and podcaster who grew up in Boca Raton, Florida. He studied computer science and economics at the University of Florida. His writing on Theo’s Substack explores artificial intelligence, technology, business, history, and philosophy. He has also contributed to Quillette and received an Emergent Ventures fellowship in 2024.

The podcast covers AI’s accelerating capabilities, broader societal impact, and safety risks. Jaffee talks about estimates of a 10% probability of human technological civilization collapsing within ten years, distinguishing between the disempowerment of humanity and the potential continuation of civilization under AI control. But, he notes, coordination, political intervention, and the logistical difficulty of automating physical infrastructure can mitigate that catastrophic risk. Progress remains broadly on track with AI 2027 expectations, despite there being no predicted labor disruptions and Chinese consolidation failing to materialize as forecast. Jaffee notes events like the Hugging Face incident and an apparent Millennium Prize problem solution heighten concerns surrounding alignment and control. Razib and Jaffee also discuss how critical uncertainties persist over how rapidly recursive self-improvement might yield superintelligence, whether mounting political backlash will establish viable safeguards or trigger heavy-handed stagnation, and how society must ultimately evaluate AI consciousness and moral status.

Subscribe now

Give a gift subscription

Share

Today Razib talks to Scott Greer, an American conservative writer, author, and podcaster who focuses on nationalism, immigration, online political culture, and the transformation of the Republican Party. Formerly an editor and columnist at The Daily Caller, he now hosts the Highly Respected podcast and newsletter and contributes to The American Conservative. He is the author of No Campus for White Men (2017) and Whitepill: The Online Right and the Making of Trump’s America (2026), a history of how paleoconservatism, libertarianism, the alt-right, and other internet subcultures helped shape Trump-era politics.

The first half of the conversation focuses on Whitepill, as Greer outlines a narrative history of the rise of the New Right and its relationship to MAGA and Donald Trump over the last decade. The conversation goes back even further, to the origins of the dissident Right in the Ron Paul movement of the late aughts. Then they move to discussing a piece recently posted on Highly Respected, Sunbeltism Takes Over America–And The South. A native of Tennessee, Greer talks about the implicit conservatism of the ascendant Sunbelt culture, but its broader cultural and social influence, and its difference from earlier Southern subcultures fixated on the “Lost Cause” and rooted in an agrarian society.

Subscribe now

Share

Give a gift subscription

On this episode of Unsupervised Learning Razib talks to Brian Chau. A graduate of the University of Waterloo and former software engineer with a background in pure mathematics, today Chau is CEO of Effort.News. He was previously with The Alliance for the Future.

Most of this conversation is around why it took so long to put together something like Effort.News, and what they’ve discovered (e.g., “The 19th-century Family Fortunes Funding Degrowth”). Chau also discusses their recruitment and capital needs as a seed-stage startup.

Subscribe now

Give a gift subscription

Share

Transcript

Transcript (PDF)

Today Razib talks to Kyle Orton. Orton is a freelance researcher, writer, and editor specializing in security and foreign policy, with a particular focus on the Middle East, the Syrian war, regional geopolitics, and terrorist movements including the Islamic State (ISIS), Iran’s Islamic Revolutionary Guards Corps (IRGC), and the Kurdistan Workers’ Party (PKK). His work also ranges widely across history, including early Christianity, Islam and its relationship with the West, intelligence and espionage, the Reformation, the English Civil War, Tsarist Russia and the Soviet Union, the First World War, Communism and the Cold War, and pre-Revolutionary Iran. Orton writes the newsletter It Can Always Get Worse, where he covers contemporary security issues alongside history, religion, intelligence, books, and occasional film reviews. His writing has appeared in publications including The New York Times, The Wall Street Journal, UnHerd, Newsweek, The Independent, The New Arab, and International Business Times.

Orton and Razib’s conversation begins with his post, The Sojourn of Arculf: An Eyewitness to “Early Islamic” Jerusalem - A Frankish bishop sheds some light on a historical dark age. Their discussion continues with probing the possibility that Islam as we know it truly did not exist any recognizable form before the 690s AD, and that the early “Saracens” as their Christian enemies referred to them were in fact a schismatic group of monotheists whose beliefs were adjacent to that of the Jews. Orton also discusses the role of Muhammad, or rather, the lack of the role of Muhammad, in the early years of the Islamic Caliphate. They also explore the possibility that later commentators largely attributed the negative reputation of Islam’s first dynastic Caliphate, the Umayyads, to the reality that they would not recognize them as Muslims. Finally, Orton talks about the reality that some of the layers of text within the Koran seem far older than the time of Muhammad, perhaps by a century or so.

Subscribe now

Give a gift subscription

Share

Today Razib talks to Gabriel Rossman, a sociologist at UCLA. The author of Climbing the Charts: What Radio Airplay Tells Us about the Diffusion of Innovation, Rossman takes a deeply analytic toolkit to questions such as why 2005’s “My Humps” became a viral hit.

Razib Khan and Rossman discuss the scandal surrounding Dr. Jason Arday, a sociologist of education at Cambridge, who was exposed for plagiarism and fabricating his credentials. Arday claimed to have run 30 marathons in 35 days and raised $5 million, which were later found to be false. The conversation highlights the poor quality of his research, which relied heavily on autoethnography, a method criticized for its subjectivity. They also discuss the broader issues of ideological capture in academia, particularly in sociology, and the challenges of maintaining academic integrity amidst political biases. The discussion centers on the perception and reality of sociology, particularly the impact of political ideologies on academic engagement. Rossman, a conservative sociologist, notes the open-mindedness of most sociologists and the importance of apolitical basic science.

Subscribe now

Give a gift subscription

Share

Human evolutionary genomics is going through the best of times and the worst of times. We know so much more than we did at the turn of the century, but we are so much less certain about the broad overall narrative of our species’ genesis within and without Africa. An elegant and legible story has given way to complexification; the scientific preference for parsimony has had to yield ground out of necessity to the simultaneous march of ancient DNA and computational genomics in the last 25 years. The elegant house of “Out of Africa,” constructed in the late 20th century via the synthesis of paleoanthropology and molecular genetics, is now looking a bit worn and shabby a generation into the new century. But there won’t be any tear down here; there’s nothing to replace our current understanding. We face an impasse as we grasp at novel findings and probe into the dark with uncertainty.

Today, our understanding of human evolutionary origins is undergoing repeated revisions and renovations. New wings are being added helter-skelter to an already sprawling edifice; alternative paths are being laid down by the startling findings of ancient DNA, with a fresh dab of theoretical paint here and there to cover up incongruities. Something new is coming into shape, but it is yet incomplete, and so ‌definitive answers elude our grasp. A host of powerful new techniques are bombarding the old orthodoxies with result after result, modification, update, and iteration. Now researchers have begun questioning exactly how well they understand the correct shape of the past that they once opined on with such certainty to the public. Over and over, the simple power of ancient DNA has thrown a wrench into the consensus models and established orthodoxies. A generation ago, we imagined that Homo sapiens, “thinking man,” emerged fully formed in Africa over 100,000 years ago and swept away all our monstrous kin before us through dint of our sheer genius. Today, it seems more likely that it was we who were the monsters out of the dark, the demons about which Neanderthal mothers would tell their little-ones.

In 2002’s The Dawn of Human Culture, Stanford paleoanthropologist Richard Klein presented what at that time was the standard model of the recent origins of humanity. Some time before 50,000 years ago, a new form of human arose through some sort of mutational jump. Klein posited that our genius, our superiority, was because of a macromutation that enabled us to generate fully articulate language. Before the emergence of this new species of human, there were many varieties of human, or, more precisely, hominin. Neanderthals in Europe and Central Asia, various “archaic” lineages in eastern Eurasia, and also descendants of other Homo forms in Africa. Modern humans, whom many popular slogans would refer to as “Africans,” rapidly swept away all these varieties of humanity after 50,000 years.

And so it was for a decade. There were dissents; in 2006 Jeffrey Wall and Michael Hammer published Archaic admixture in the human genome. This paper reflected suspicion among many evolutionary geneticists that the orthodoxy promulgated was too pat and simplistic (see also Magnus Nordborg’s 1998 On the Probability of Neanderthal Ancestry). But‌ despite an underground counter-consensus, very few evolutionary geneticists were vocal on this issue in public, no matter the private grumblings they may have had (Gregory Cochran and Henry Harpending’s 2009 book, The 10000 Year Explosion: How Civilization Accelerated Human Evolution, was the exception to the rule). This all changed in 2010, when Svante Paabo and colleagues reported that Neanderthal whole genomes yielded strong evidence of several percent admixture into non-Africans, as well as the discovery of a new human lineage in eastern Eurasia, the Denisovans, who also contributed about 5% of the ancestry of Papuans.

Previous work had almost entirely been a matter of inference derived from contemporary populations. You looked at genetic variation in people alive today and worked backward to plausible models of how that variation could have arisen. In the 1980s, geneticists examined mitochondrial lineages, which represent the direct matrilineal genealogy. Geneticists noticed that all non-African populations nested within African genetic variation. They soon replicated this result with the Y chromosome, passed only through males, and the autosomes (markers on chromosomes 1 through 22) representing the whole genetic heritage. These results neatly dovetailed with the findings of paleoanthropologists like Chris Stringer of the British Museum of Natural History, who argued that modern human morphology, mostly exemplified by traits in human skulls, reflected continuity with African Homo, and not Neanderthals. Following the molecular genetic results, researchers applied similar phylogenetic methods to morphometric traits and discovered the same pattern of non-Africans nesting within African variation.

It was an immaculate and tidy story. The broader public became very familiar with this narrative not just because of the scientific literature, but through documentary series like Nova’s 1987 Children of Eve or Spencer Wells’ 2000 Journey of Man, the latter of which was viewed by tens of millions. It’s no surprise that the first ten minutes of Journey of Man lead up to Wells boarding a flight to Africa, where he reassures us that the human story began.

And that still seems to be much of the story. But not the entire story. Over the last few years, geneticists have concluded that a much higher fraction of non-African DNA was originally Neanderthal. In an interview with Dwarkesh Patel Harvard’s David Reich asserted that as much as 10-20% of the overall heritage of early non-Africans, just as they were expanding out of the Near East 50,000 years ago, may have been Neanderthal. How then is that today we detect only about 2% Neanderthal genes outside of Africa? The genome rapidly sheds genetically incompatible segments within a few thousand years via purifying selection. Because Neanderthals diverged from our predominantly African ancestors 600 to 700,000 years ago, their overall genetic makeup exhibited much more striking incompatibilities with the expanding Africans than occurs when different branches of our own species mix (the deepest division in our own lineage dates to about 200,000 years, when Khoisan ancestors diverged from everyone else). No doubt the same phenomenon applied to Denisovan admixture, which today in some Oceanian populations, like those in New Guinea or the indigenous populations of the Philippines, approaches 5% or so.

What does this all mean? Because natural selection changes allele frequencies in ways that are out of step with the overall genome, the signatures that we get from modern and ancient DNA are deceptive as to the demographic dynamics of early anatomically modern humans and the Neanderthals (and Denisovans) whom they encountered. All the evidence, both ancient and modern, points to a tiny non-African ancestry population between 50 and 60,000 years ago, a few thousand individuals at most (some models posit a bottleneck of 200 breeding individuals!). If 10-20% of the ancestry of the early modern human expansion, also known as the Initial Upper Paleolithic (IUP), was Neanderthal, that implies the integration of hundreds of Neanderthals, as opposed to ten or twenty. What might the cultural implications of this be? Perhaps early modern humans integrated some Neanderthal folkways, and even loan-words from the Neanderthal ur-language? The reduction of Neanderthal ancestry was a genomic-scale process that took approximately 1,000 years, during which cultural exchange could still have occurred in the mixed individuals.

As I note in this monologue, I am sitting on a podcast previously recorded about Jason Arday because of his recent death. Instead, I’m posting my thoughts about what happened before he died, and since then, and how it relates to Nathan Cofnas.

Links:

• DEI Fraud and Cover-Up at Cambridge-University refuses to investigate DEI star’s plagiarism

• Institutional Failures in the Garden of Great Intellects

• Professor’s Death Leaves Behind Hard Questions for Cambridge on Race and Hiring

Subscribe now

Give a gift subscription

Share

Nikolai Yakovenko is back on the podcast because of the Hugging Face hack. Yakovenko is a former professional poker player, and was a research scientist at Google, Twitter and Nvidia. With a decade in computer science, Yakovenko has been at the forefront of the large-language-model revolution that has driven to prominence companies like OpenAI, Anthropic, and DeepMind, and an ecosystem that has birthed hundreds of smaller startups. He is also the founder of DeepNewz, an AI-driven news startup. His latest startup is HuggingNews.

Razib and Yakovenko talk about the implications of OpenAI’s inadvertent hack of Hugging Face, and whether he has been too sanguine about the possibility of AI-apocalypse. Though Yakovenko hasn’t gone full doomer, he admits that this sort of behavior by AI agents is alarming. That being said, he cautions we need to keep perspective. They also discuss predictions of the coming obsolescence of software engineers…Yakovenko notes that in some ways Dario Amodei was correct to predict the end of software engineering as a profession: very few write code unaided by LLMs today. In a few years, many professions, especially in tech, have transformed, but warnings of mass unemployment have not materialized.

Subscribe now

Give a gift subscription

Share

Today, Razib talks to internet personality Werner K. Zagrebi. A statistics and machine learning student at Carnegie Mellon University, Zagrebi has led a peripatetic existence over the past few years, from being Curt Yarvin’s personal assistant and editor of Gray Mirror, and working to launch a media company and think tank . Of late, he has been posting …

Today, Razib talks to Collin Hogue-Spears, a cybersecurity and artificial intelligence governance expert with more than 20 years of experience in technology, product management, federal compliance and software security. Hogue-Spears is an independent researcher and the author of From Lab to Life: How AI Works in China. His analysis has appeared in publications including The Wall Street Journal and Politico.

Razib and Hogue-Spears first talk abut China’s rapid AI development, driven by provincial initiatives and significant investments in infrastructure. Hogue-Spears emphasizes China’s focus on practical AI applications, contrasting it with the U.S.’s emphasis on AGI. He predicts China’s dominance in AI, particularly in embodied AI, and warns of the potential global impact if China surpasses the U.S. in AI development. The conversation highlights the striking differences between the U.S. and Chinese technological ecosystems.

Subscribe now

Share

Give a gift subscription