Razib Khan's Unsupervised Learning

In June 1991, The New York Times published a piece titled “Scientists Study Ancient DNA for Glimpses of Past Worlds.” Published a year after Michael Crichton’s 1990 novel Jurassic Park, on which the 1993 blockbuster would be based, the article opens “Will it one day become possible to breed a living dinosaur from genes preserved in fossils?” More than 30 years on, we obviously have not bred a living dinosaur, nor come even close. But the early 1990’s kicked off the first age of ancient DNA with massive optimism, stimulated by the spread of polymerase chain reaction (PCR) technology, which allowed the amplification of minute amounts of genetic material. I  remember this particular article because an extract was passed around my 8th-grade science class. Today I talk to Dr. Elizabeth Jones, author of Ancient DNA: The Making of a Celebrity Science, about the excitement in paleogenetics in the 1990s, its subsequent retrenchment as a field, and the prompt renaissance in the late 2000’s under the leadership of Svante Pääbo and Eske Willerslev.  Jones first articulates her idea of what “celebrity science” is. Rather than science being driven by an individual, like Albert Einstein or James Watson, she argues that fields of science itself can become a celebrity due to strong public interest. This was clearly true in the 1990’s, with the frequent popular press stories about dinosaur DNA and the tie-in to Jurassic Park, and in the late 2000’s and down to the present, culminating in Pääbo being awarded a Nobel Prize. In her book, Jones views celebrity science as a wholly positive development, allowing for a full exploration of the possibilities of a field, and putting the spotlight on possible problems so that science’s self-correction mechanisms can kick-in. This was certainly true with the first “hype cycle” of ancient DNA, when it turned out that many of the specific results were actually the result of modern contamination. Jones outlines how in fact it was Pääbo and his collaborators who were the skeptics demanding stricter methods and frameworks before publication. Ancient DNA then recounts how a second hype cycle was triggered by genomic sequencing’s integration into paleogenetics, opening up an avenue to obtain the whole genomes of Neanderthals and other species. Jones’ Ph.D. work, which led to Ancient DNA, was just concluded in 2015. But she recalls the mad last-second scramble to update the manuscript at the last second because of this last half a decade’s flood of results. They consider the field’s future possibilities, and finally the prospects for genuine retrieval of biological material from actual dinosaurs, the original ambition in the 1990’s

On this episode of Unsupervised Learning, Razib talks to journalist Lee Fang. Formerly an investigative reporter at The Intercept and a contributing writer at The Nation, Fang began his journalism career at ThinkProgress. As an undergraduate, Fang was president of the University of Maryland College Democrats, and interned for Democratic representatives Stephanie Tubbs Jones and Steny Hoyer. He was also the first intern for the progressive media watchdog group Media Matters for America. Today Fang publishes his findings on his eponymous Substack. Razib asks Fang what it’s like to do investigative journalism without institutional support. A common assertion in relation to the decline of the mainstream media is that only organizations with deep pockets and long time horizons can fund such work, but Fang has already exposed Pierre Omidyar’s support for defunding the police while personally investing in private security companies, discovered that a major California diversity consultant was defrauding the state and found that tech companies put in mass applications for lower-paid H1-B hires after mass layoffs of mostly Americans. Fang attests to there being a surfeit of public documents there for the taking; what's needed is just the will and determination to dig. Razib and Fang also discuss what it’s like to start out in progressive journalism, but now find yourself in an ideological space without a clear partisan valence. Though Fang clearly comes out of the Left, some of his findings, like that Pfizer funded lobbying groups who pushed vaccine mandates, are now more palatable to the Right. He is also unafraid to take on establishment progressive pundits, like Mehdi Hassan, whom Fang systematically exposed as having a history of engaging in personal attacks based on accusations of racism as well as engaging in plagiarism.

On this episode of Unsupervised Learning Razib hosts Ross Douthat, author of Bad Religion: How We Became a Nation of Heretics, Privilege: Harvard and the Education of the Ruling Class, Grand New Party: How Republicans Can Win the Working Class and Save the American Dream, The Deep Places: A Memoir of Illness and Discovery and The Decadent Society: How We Became the Victims of Our Own Success. A columnist at The New York Times, often on political and social topics, Douthat also reviews movies for National Review. Today Razib talks to him about a topic that is a bit off the beaten path: why is genre fiction, and fantasy literature, still relevant today, and how, in the last generation, did it break out of its cultural ghetto? First, Douthat addresses the massive role that J. R. R. Tolkien’s works have had on the field, how most of modern fantasy is either an imitation of his works or a response to them, and the creative limitations that that circumstance imposes. Razib and Douthat then discuss the various shifts in the genre style over the last 20 years, toward a more gritty and morally ambiguous style exemplified by George R. R. Martin, and the cultural breakthrough of Game of Thrones in the 2010’s. Douthat avers that in some ways genre has come into its own, with the decline in the cultural status of realistic fiction and drama, and the ascension of “comic book movies.” While Razib believes that Marvel films are arguably fantasy, Douthat disagrees, believing their contemporary settings disqualify them. They also address whether fantasy is actually simply the channeling of premodern narrative styles that go back to the Epic of Gilgamesh or the Iliad. They address the somewhat exceptional success of Game of Thrones on television and the failure of Amazon’s Rings of Power and The Wheel of Time. Finally, Douthat talks about his unpublished fantasy novel, The Falcon’s Children. 

On this week’s Unsupervised Learning Razib welcomes back a favorite repeat guest, Samo Burja, to discuss matters future, present and past. Burja founded the consulting firm Bismarck Analysis and developed the “great founder theory.” He contributes to Palladium Magazine, Asia Times, City Journal, and The National Interest. Burja’s first appearance on the podcast, recorded in the fall of 2020, spiraled into a long discussion on the Chinese past and future, and Razib follows up to find out where he thinks the Peoples’ Republic is in 2023 and where it will go in the near future. Though Burja acknowledges that bullishness on Chinese power has abated, and Xi’s leadership has left much to be desired in its outcomes during the tail end of the COVID-19 pandemic, he asserts that China’s capacity as a manufacturing power cannot be underestimated. Even if the vision of a Chinese hyperpower that prevailed at the end of the 2010’s has receded, Bujra foresees a multipolar world. Jumping to the present, Razib also revisits the Russia-Ukraine war, a topic they discussed in a subsequent podcast last year. Burja notes that the rise of Ukrainian nationalism over the last few years is a somewhat unexpected and novel development, and Putin clearly made a misstep. That being said, Razib and Burja agree that the war seems likely to drag on indefinitely as neither participant has a plausible clean exit strategy. Finally, the second half of the conversation jumps back to the past, as they discuss a piece Burja co-authored for Palladium two years ago, Why Civilization Is Older Than We Thought. Here Burja believes he is now on firmer ground than in 2021, as new findings even more ancient than Göbekli Tepe have since been uncovered. Razib and Burja discuss the difficulty of navigating between the excessive conservatism of archaeological science and the flights of fantasy that some popularizers like Graham Hancock indulge in when it comes to the idea of “Ice Age civilizations.” All that being said, Burja believes it is not outside of the realm of possibility that a civilization as complex as ancient Egypt may have existed during the Ice Age.

A few years ago now, Razib talked to Tim Lee about his new Substack Full Stack Economics, which featured deep dives into economic issues (as well as some on-the-ground-reporting, like when he drove Lyft to get a feel for its economics). But recently, Lee decided to put Full Stack Economics on pause to focus on a new Substack: Understanding AI. Artificial intelligence is hot right now, but Lee covered tech for a decade for Washington Post, Ars Technica, and Vox.com, and has a master’s degree in computer science from Princeton, so Razib was curious about what he’s learned. Recently two pieces seem to illustrate the alternative faces of generative AI and LLMs, I cloned my voice with AI and my mother couldn’t tell the difference and Why I'm not worried about AI causing mass unemployment. On this episode of Unsupervised Learning, Razib and Lee discuss the impact of artificial intelligence, the good, bad and trivial. Lee makes the case that AI might be like the internet, transforming narrow aspects of knowledge-work and enabling a richer culture, but without clear revolutionary implications for the economy. His thesis hinges on the fact that AI cannot operate in the material world due to the primitive state of robotics, though Razib wonders if this barrier too might fall in the near future. Overall, Lee suggests that the AI “hype machine” is being driven by the fact that information workers who set the terms of public discussion are the ones likely to be most impacted; waiters, plumbers and nurses, in contrast, will be just fine. Razib and Lee also discuss the field of existential risk analysis. Lee found Nick Bostrom’s Superintelligence: Paths, Dangers, Strategies unimpressive due to the fiat assumption that fields like nanotechnology could be manipulated by AI when nanotechnology itself is in quite a primitive state. Overall, Lee’s perception is that the “doomers” in the AI punditry field tend to be science fiction writers who are better at spinning narratives than doing analysis.

Recently, scientists discovered that a two-year mega-drought beginning in 1198 BC hastened the Hittite Empire’s collapse. The finding sheds new light on the history of the decades around 1200 BC, adding specificity to the timing and cause for the period’s social and political chaos. Today on the Unsupervised Learning podcast Razib discusses the “Bronze Age Collapse,” the end of the first globalized world. This collapse marked the end of a multi-century period when the Near East’s empires and states had matured to the point where a common system of diplomacy and trade could emerge, centering on the three major foci of Egypt, Mesopotamia and Anatolia, but including smaller states like the kingdoms of Greece and the Levant. The end of this world was so sharp and abrupt that many regions, like Greece, lost their memory of these centuries, plunging into a “Dark Age,” when urban life disappeared and life regressed back to a Neolithic scale. The Hittites were entirely forgotten, recalled only in a few mentions in the Bible. Razib discusses whether these sorts of shifts are inevitable in any sufficiently complex social system, or whether external shocks like climate change are necessary. What can the end of the Bronze Age tell us about other civilizational collapses?

On this episode of the Unsupervised Learning podcast, Razib talks to Peter Nimitz, the author behind the Nemets Substack, which explores topics as diverse as the 2014 Donbass War and the likelihood of Eurasian migration into Chad thousands of years ago. Razib and Nimitz walk through his recent post, the Seven Ages of Western Eurasia: A brief outline of the 11,700 years from the Anatolian Farmers to the Present. In the piece, he explores the changes that Europe and West Asia have undergone since the end of the last Ice Age, including the rise and fall of pre-literate civilizations before written history, and the recurrent social and economic collapses from which humans have had to rebuild. Razib and Nimitz have similar interests, but where Razib focuses more on genetic relationships, Nimitz tends to dive deep into archaeology, supplementing his understanding of the migration of peoples with paleogenetics. They also discuss the proto-civilizations of the Ice Age, including nascent farming communities that might date to over 20,000 years ago. The conversation repeatedly circles back to the theme that paleogenetics has had a transformative effect on interpreting archaeological sites and our understanding of the migrations of past peoples. https://razib.substack.com This is where you will find all the podcasts from Razib Khan's Substack and original video content.

Alex Feinberg is anything but your typical trainer. An economics graduate from Vanderbilt, Feinberg willed himself to become a professional baseball player through focus and hard work and then talked his way into a sales and business development job at Google. In the late 2010’s Feinberg moved into the crypto space, but found that one precondition to success was having a large Twitter following. So he pivoted and focused on growing his Twitter following, and noticed that his lifestyle tweets, and images of him highlighting his fitness and dietary regime, were the ones that gained traction and got him followers. Today Razib talks to Feinberg about how he helps people optimize their health, and how that might apply to other aspects of their life, like getting a raise at work through changing your presentation and delivery. Feinberg explains his methods rely on intuition and try to leverage your cognitive biases, as opposed to working against them. Razib probes whether these methods are appropriate for everyone; perhaps there are cognitive and physical parameters that help in terms of optimization. Feinberg admits his techniques are more effective for males than females. For example, it is much easier to gain muscle mass by simply changing nutrition and lifting if you have high testosterone. The Feinberg phenotype is not accessible to all. Feinberg and Razib also talk about his experiences in the corporate world, in particular at Google. Feinberg claims that in 2011 Google was a much more libertarian company than it was by the time he left in 2018. In particular, Feinberg noticed that many executives and senior managers began to lie and willfully mislead employees in order to scale the corporate ladder. These experiences helped convince him that the corporate world was not for him, and over the last five years, his focus has been on developing independent revenue streams through his coaching, training and books.

On this episode of Unsupervised Learning, Razib talks to Adam Mastroianni, who runs the Experimental History Substack. Mastroianni was the inaugural guest on the Intrinsic Perspective podcast, hosted by Erik Hoel, where they discussed his post, The rise and fall of peer review - Why the greatest scientific experiment in history failed, and why that's a great thing (see also his follow-up, The dance of the naked emperors). Mastroianni opened a can of worms; the post has more than 800 likes and more than 330 comments. Razib asks Mastroianni about the fiercely positive and negative reactions to his contention that modern peer review has outlived its utility. They also unveil the historically contingent origins of the practice in the mid-20th century, and how it came to be seen as a holy enterprise necessary to science. Both agree that scientific publishing needs a paradigm shift; a topic that Razib tackled in 2014 with the Genome Biology comment Dragging a scientific publishing into the 21st century. Razib and Mastroianni then discuss Experimental History, a Substack devoted to social psychology and meta-science. Why has Mastroianni decided to devote a substantial amount of energy to this project, as opposed to just publishing in journals?  Experimental History touches on some of the experimental social psychology research Mastroianni has been involved in, but it also focuses on some of the generally understood findings in psychology and neuroscience, and why they’re true or false. In a world of academic science saturated with Ph.D. level researchers, Razib and Mastroianni explore the communication possibilities inherent in the Substack model. Finally, Mastroianni unpacks his opinion that even many of the robust statistically significant findings in social psychology don’t matter. He believes that the lack of a single theory blocks proper understanding in psychology, and many of the results in his field are both uninteresting and fail to lead to a nontrivial increase in knowledge.

This week on Unsupervised Learning, Razib and his guest, David McKay, of the Standing on the Shoulders of Giants podcast (Razib was an early guest), discuss the rise of artificial intelligence (AI) and the prospects for artificial general intelligence (AGI). This discussion arose after Razib heard McKay’s explainer, Zen and the Art of ChatGPT, a 30-minute layman’s intro to the topic, where he breaks down the technical elements that come together to allow for AI. In this episode, McKay, a Cambridge University-trained computer scientist who has worked at Hotmail and Google, digs deeper into the nature of Large Language Models (LLMs) and how they give rise to probabilistic generative AI like ChatGPT and whether we should be worried. Razib’s conversation with McKay follows another recent episode on AI. I  the earlier podcast, Nikolai Yakovenko: GPT-3 and the rise of the thinking machines, the interviewee, a computer scientist, was relatively sanguine about the world-ending possibilities of AGI. McKay generally takes the same position, highlighting the reality that most computer scientists and AI researchers are less worried about science-fictional apocalyptic scenarios than the general public or AI-skeptics like Eliezer Yudkowsky and Nick Bostrom (the author of Superintelligence: Paths, Dangers, Strategies) are. And yet the reason that AI is so topical is it seems that the development of the technology is proceeding along an exponential path; ChatGPT 4 was released months after ChatGPT 3. McKay and Razib also discuss the release of Bard, Google’s chatbot, and the offering from Microsoft’s Bing, and how they are similar and different from ChatGPT.  While McKay is optimistic about the possibilities of AI as a tool, ultimately, he is in the camp that believes it really isn’t intelligent in the same way as a human. Because it relies on the corpus from the internet, ChatGPT cannot really do math. It lacks true conceptual understanding that would allow it to grasp truth beyond what the internet might tell it. Razib and McKay also talk about the energetic resources that LLMs consume (Microsoft had to reallocate compute resources after the release of Bing’s chatbot), and how that might be a limitation on their scalability.