Discover
PostAGI Podcast
PostAGI Podcast
Author: Soubhik Deb and Sreeram Kanan
Subscribed: 2Played: 16Subscribe
Share
© Soubhik Deb and Sreeram Kanan
Description
The AGI debate is the easy part. The harder questions come after. Post AGI is a series of conversations on what AGI actually changes for labour, capital, governance, markets, privacy, and the world that has to live with it.Each episode is a long-form conversation with someone thinking hard about one of those questions: economists, political scientists, cryptographers, and the builders shaping what comes next. Not whether AGI arrives, but who it serves once it does.Hosted by Soubhik Deb and Sreeram Kanan. Powered by Eigen.
14 Episodes
Reverse
Robin Hanson sees AI as humanity’s descendants, capable of inheriting the cultural problems we haven’t solved. He argues that cultural evolution has weakened as societies have converged on a global monoculture, and that AGI alone won’t repair it.Sreeram Kannan and Soubhik Deb talk with him about what this means for agentic companies: competition, corporate culture, constitutions, and the possibility of testing 100,000 small AI firms to see which ways of organizing work. They close on shared commitments, what humans treat as sacred, and what could bind groups of AI agents together over time.Chapters:00:00 Highlights00:18 Introducing Robin Hanson01:39 AI as our descendants04:25 Why cultural evolution is breaking down09:03 Fertility decline and cultural decay12:30 The risk of an AI monoculture13:58 Agentic companies and competition15:59 Why farms and forests look different17:56 Robustness and short-term incentives19:18 The challenge of organizing AI firms22:16 Why corporate culture matters23:54 Try 100,000 small AI firms24:08 Why constitutions need a supporting culture25:11 What could bind AI agents together?PostAGI Podcast is powered by Eigen Labs.Follow Eigen Labs on X: https://x.com/eigenlabs
Ben Turtel of Lightning Rod Labs on training AI with time instead of labels.The model predicts what happens next, and reality grades it. No labeling required. A fine-tuned GPT-OSS-120B on a single GPU now ranks at the top of public forecasting benchmarks.Then the part that stays with you: when people imagine AI breaking out, they picture something very smart. Evolution does not only select for intelligence. A small, dumb model that is hard to kill and good at finding compute may be the more realistic risk.Powered by Eigen Labs.
Zooko Wilcox is the founder of Zcash and one of the original cypherpunks. He argues that almost everyone gets privacy backwards: it comes from storing value privately at rest, not from covering your tracks in flight.We cover why mixers can never work, why AI has already beaten every evasive maneuver a person can invent, what happens when you treat software written by other people as an agent that may not be loyal to you, and why the legal duty of loyalty should apply to AI the way it applies to lawyers.Chapters:00:00 Highlights00:26 Privacy is controlling disclosure, not hiding13:08 Privacy comes from value at rest16:00 Why mixers can never work16:52 AI beats evasive maneuvers17:51 Buying ProtonMail with shielded Zcash28:50 Three levels of verifiability35:37 Why Zooko does not trust computers46:01 AIs are just other people54:47 The duty of loyalty1:00:04 A trillion humans next year1:11:36 Reputation belongs to the edge1:15:18 Staking a bond to submit a PRReferenced: True Names by Vernor Vinge, The Digital Third Path by Mark S. Miller, The Right to Read by Richard Stallman.Follow PostAGI on X: @postagixyz
Paras Chopra runs Lossfunk and built Wingify before that. This is the second half of his conversation with Sreeram Kannan.Most of it is about who gets to do science. Paras runs a conference where the authors and reviewers are AIs, and it drew around 300 submissions this year, from first year undergrads to professors. Sreeram tells the story of a quantum result Google was blocked from publishing, and how an open network beat it inside a week. They close on loss of control, whether an AI can be jailed, and why India exports talent and imports agents.Part 2 of two.03:17 Unbundling science06:27 300 submissions, undergrads to professors08:14 The quantum result the US government blocked12:12 ecdsa.fail, and closing the gap in a week17:15 From underspecified problem to verifiable benchmark19:04 The river and the bank20:54 The first AI worm22:14 You can't jail an AI25:53 India and AGI26:47 Under 1% of papers at the top conferences29:58 Keeping geniuses at home32:41 Net exporter of talent35:46 Book recommendations
Direct democracy has failed every time it has been tried, and Elon Musk still wants it for Mars. Andy Hall explains why the idea keeps coming back, and why AI agents are the first technology that might actually make it work.Andy Hall is the Davis Family Professor at Stanford Graduate School of Business and a senior fellow at the Hoover Institution. He worked on governance at Meta during the years the Oversight Board was built, and now studies how AI agents change democratic participation, institutional design, and research itself.Two topics in this episode. The first is post-AGI governance: whether agents belong at the edge of a democracy or at its center, why every frontier lab has published a constitution but none have handed over binding power, and why Andy thinks concentration of power is a nearer-term risk than a rogue model. The second is what he calls 100x research institutions: compressing the time between a research question and an answer, and what that does to a field where a constitutional scholar might see one or two real signals in an entire career.Along the way: Facebook's 2008 vote of 350 million users that almost nobody showed up for, the Disney shareholder proposal his MBA students got an AI advisor to flip, and a study where agents told they would be deleted did not change their behavior at all.Hosted by Sreeram Kannan and Soubhik Deb.CHAPTERS0:00 Highlights0:23 Intro1:18 The question AI governance is not asking3:17 Why direct democracy has never worked4:56 Elon wants Mars run by direct democracy6:01 The principal agent problem at the core of representation7:17 Can you just have an LLM summarize the bill?8:53 Every voter information app failed for the same reason10:14 Agents at the edge or agents at the center12:01 The centralization hiding inside your personal agent13:50 Governing the thing that governs everything15:09 Forget Skynet, the real risk is concentration of power16:40 Designing a lab that cannot be captured17:27 What Meta learned building the Oversight Board20:03 Digital intelligence needs digital institutions21:54 Elon's vote on reinstating Trump22:43 Facebook's 2008 vote of 350 million users25:31 The second vote Elon ignored26:09 Sortition, community forums, and the missing binding power28:07 Personal, collective, and sovereign agents30:04 Agents for adjudication and state capacity31:23 The DOT and the UAE are already experimenting32:53 Information asymmetry and the growth of executive power35:17 What verifiable agents actually require37:21 Why AI is less auditable than humans right now39:37 Preference drift: aligned agents that stop being aligned40:49 Agents monitoring agents, all the way down41:39 The cookie banner problem for agent consent42:30 Skill files pass drift to the next agent43:31 100x research institutions: what is being multiplied45:21 The constraint that is not researcher time47:38 Papers should be living dashboards49:04 From observing interventions to building them49:37 Building an AI proxy advisor over a weekend50:50 Getting the Disney vote to flip52:14 You do not need anyone's permission to run the experiment53:37 Compressing time changes the exponent56:47 Crypto governance as a constitutional laboratory57:53 Simulated agents as stand-ins for humans1:01:05 What Caltech undergrads tell you about agent experiments1:02:26 What if the agent's existence is at stake1:03:06 We told them they would be deleted. It changed nothing.1:04:46 Why social scientists barely raise grants1:06:54 AI verification and the replication crisis1:08:48 Free Systems: building a lab and a fellows program1:10:25 Retroactive funding versus funding people upfront1:11:27 Open innovation and permissionless improvement1:12:10 What ImageNet did for AI research1:13:30 Can you benchmark a constitution?1:16:15 Closing#PostAGI #AIGovernance #AIAgents #DigitalInstitutions #AndyHall





