No Priors Ep. 138 | The Best of 2025 (So Far) with Sarah Guo and Elad Gil

Detailed Summary

Episode Introduction (0:00 - 0:21)

The episode introduces itself as a compilation of favorite moments from the No Priors podcast in 2025, highlighting remarkable advancements in AI and featuring discussions with leaders from companies like Harvey, OpenAI, Glean, Abridge, and scientific legends like Dr. Fei-Fei Li and Noubar Afeyan.

Winston Weinberg on Leaning into New Capabilities (0:21 - 2:01)

• Winston Weinberg, CEO of Harvey, recounts discovering the potential of GPT-3 for legal workflows when it was not widely recognized.
• He and his colleague Gabe tested GPT-3 on 100 landlord-tenant questions from Reddit's r/legaladvice.
• They presented the AI-generated answers to three landlord-tenant attorneys without mentioning AI, asking if they would send them to clients without edits.
• A remarkable 86 out of 100 answers were approved, leading to a meeting with OpenAI's C-suite after their general counsel expressed surprise at the model's legal proficiency.

Dr. Fei-Fei Li on Spatial Intelligence (2:01 - 4:13)

• Dr. Fei-Fei Li discusses spatial intelligence as a fundamental and challenging problem that evolution solved for animals, particularly in reconstructing a 3D world from sensory input.
• She highlights that even for humans, generating complex 3D models from memory is difficult without specialized training, indicating it's not a fully solved problem.
• The ability for AI to easily and fluidly interact with and edit 3D environments would open up a "whole different world."

Brendan Foody on AI Disruption in the Workforce (4:13 - 6:10)

• Brendan Foody, CEO of Mercor, predicts rapid and painful workforce displacement due to AI, leading to significant political problems and populist movements.
• He emphasizes the critical need to address how society will reallocate wealth and redefine roles as AI approaches superintelligence.
• Foody suggests displaced workers might transition to roles in the physical world, such as robotics data creation, service industries (e.g., waiters), or therapy, where human interaction is valued.
• He believes automation in the physical world will progress slower than in the digital world due to self-reinforcing gains and self-improvement loops inherent in virtual environments.

Dan Hendrycks on the Geopolitics of Superintelligence (6:10 - 8:06)

• Dan Hendrycks, Director of the Center for AI Safety, draws an analogy between superintelligence and nuclear strategy, where shared vulnerability deterred first strikes.
• He suggests that as AI becomes pivotal to national futures, states will deter each other from leveraging it as a "super weapon" or for rapid, automated AI research and development loops.
• Hendrycks foresees potential preemptive cyberattacks on data centers by nations like China and the US to prevent rivals from gaining an AI advantage, with other nations like Russia reassessing their positions.

Noubar Afeyan on Entrepreneurship (8:06 - 10:38)

• Noubar Afeyan, founder of Flagship Pioneering and co-founder of Moderna, discusses his motivation to transform entrepreneurship from a random, improvisational, and "gamy" process into a scientific endeavor.
• He recounts starting a company in 1987 as a young immigrant, a rarity at the time, and questioning why entrepreneurship wasn't treated as a profession.
• Afeyan criticizes the perception of entrepreneurship as a game with winners and losers, where failure is expected, especially when dealing with significant investments and critical areas like healthcare, climate, and food security.
• He advocates for a more systematic approach, believing that the process can be improved and made more predictable.

Brandon McKinzie and Eric Mitchell on Reasoning Models (10:38 - 12:41)

• Brandon McKinzie and Eric Mitchell from OpenAI explain that reasoning is a major paradigm shift in AI architecture since the transformer.
• They provide concrete examples, such as visual reasoning models transparently admitting uncertainty and then using tools to manipulate images for better understanding.
• This tool-use significantly improves the model's efficiency and performance, leading to a much steeper test-time scaling slope.
• Similarly, for tasks like writing code or performing complex calculations, an LLM can either attempt self-verification (which is slow) or write simple programs to achieve verifiable and faster results, deferring tasks to tools where they have a comparative advantage.

Isa Fulford on Training Deep Research (12:41 - 13:49)

• Isa Fulford from OpenAI describes the "visceral experience" of AI development, where the most profound moments come from seeing something hoped for actually work.
• She highlights the surprise and satisfaction of training a model on a new dataset, like browsing tasks, and witnessing it perform effectively, even when there was prior conviction it would work.
• Fulford also notes that the model's unexpected mistakes, despite its overall intelligence, indicate significant room for improvement.

Arvind Jain on Innovating Enterprise Search (13:49 - 16:21)

• Arvind Jain, founder of Glean, explains how AI has transformed enterprise search from a "graveyard market" into a viable solution.
• He attributes past failures to the difficulty of accessing data in pre-SaaS environments.
• The advent of SaaS systems, with their standardized versions, interoperability, and API access, solved the biggest problem: bringing all enterprise information into one place for unified search.
• Jain founded Glean after experiencing this problem firsthand at Rubrik, where employees struggled to find information across 300 SaaS systems.
• He notes the massive explosion of content within businesses, with one large customer having over 1 billion documents, a quantity comparable to the entire internet in 2004, necessitating scalable cloud-era systems.

Dr. Shiv Rao on AI’s Human Impact (16:21 - 18:58)

• Dr. Shiv Rao, CEO and founder of Abridge, shares how AI is making a profound human impact in healthcare.
• He describes receiving feedback from doctors, both critical (which is valued as "oxygen") and overwhelmingly positive, expressing renewed career longevity thanks to AI technology.
• Rao recounts a particularly moving "love story" from a doctor at Tanner Health, a rural system, who explained to her son that Abridge allows her to come home early for dinner every night.
• This type of feedback provides a deep sense of purpose and fulfillment for the Abridge team, highlighting the "oxytocin hits" that sustain them through the pressures of building new technology.

Conclusion (18:58 - 19:00)

The hosts conclude by emphasizing that these conversations underscore a "hinge moment in history" and encourage listeners to stay tuned for more insights from AI builders and thinkers.

Notable Quotes

"Oh, I had no idea the models were this good at legal." — OpenAI General Counsel

"I think displacement in a lot of roles is going to happen very quickly and it's going to be very painful uh and a large political problem." — Brendan Foody

"Let's think of what happened in in nuclear strategy. Basically, a lot of a lot of states deterred each other from doing a first strike because they could then retaliate. So, they had a shared vulnerability." — Dan Hendrycks

"Why isn't entrepreneurship a profession? And if it was going to be a profession, how could it be a profession?" — Noubar Afeyan

"It really was one of those things where we thought that, you know, training on browsing tasks would work... But actually the first time you train a model on a new data set using this algorithm and seeing it actually working and playing with the model was pretty incredible." — Isa Fulford

"It was like a graveyard like you know of all these companies that tried to solve the problem and it didn't." — Arvind Jain

"I was sitting at dinner last week and my son asked me mommy why aren't you working right now I literally took my phone out and explained to him that a bridge is a new tool that lets mommy come home early and eat dinner with her family." — Doctor at Tanner Health