Detailed Summary
The hosts, Sarah Guo and Elad Gil, reflect on the rapid advancements in AI during the past year, noting its mainstream adoption and impact on policymakers. They highlight key developments such as:
- ChatGPT's massive growth and financial demands.
- Google's Gemini roaring back strong.
- AI coding shifting to agents, consuming inference capacity.
- Widespread adoption of clinical decision support by doctors.
- Accelerated enterprise adoption in legal and customer support.
- The emergence of new research labs and a focus on diverse research directions like diffusion, self-improvement, and data efficiency.
- Renewed optimism in robotics, with companies expected to face real-world deployment challenges.
- A personal prediction that LLMs will enable significant financial gains in market trading.
Elad Gil outlines several key predictions for 2026, emphasizing the cyclical nature of AI hype and the continued, understated progress of the technology:
- Expectations of renewed "AI bubble" claims and arguments that AI isn't delivering, despite the technology's fundamental value.
- Prediction that the next set of verticals will achieve massive scale, following the consolidation seen in coding, medical scribing, and legal AI.
- Sarah Guo notes investor stress regarding capital deployment and uncertainty around adoption cycles, but acknowledges the blinding speed of overall AI adoption.
Adoption of AI in Professional Fields (04:40 - 07:17)
The discussion highlights the surprising and rapid adoption of AI in traditionally conservative professional sectors:
- A report from "Off Call" indicates significant AI adoption by doctors for documentation and clinical decision support.
- Elad Gil points out the fascinating trend that physicians, lawyers, accountants, and compliance professionals—historically slow tech adopters—are now embracing AI quickly.
- Sarah Guo agrees that this trend will continue, as AI's ability to reason and interact with unstructured data is highly valuable in these fields.
- They acknowledge that market fluctuations (e.g., Nvidia's performance) might cause short-term panic but won't deter the fundamental secular change driven by AI.
- Elad Gil predicts the emergence of new foundational models focused on scientific progress in physics, materials science, and mathematics, leading to anecdotal breakthroughs that will be initially overhyped but ultimately profoundly impactful.
Robotics and Self-Driving Cars (07:17 - 13:59)
The conversation shifts to the future of robotics and self-driving technology, including a debate on timelines and market dynamics:
- Sarah Guo predicts a "collapse of sentiment" around some robotics companies in 2026 due to unmet timelines, despite overall field progress.
- Elad Gil believes humanoid and semi-humanoid robots will be deployed at small scale in consumer or industrial environments, leading to a "freak out" when they don't perfectly work.
- They draw parallels to the 15-17 year journey of self-driving cars, suggesting robotics will follow a similar, albeit potentially faster, curve.
- A key question is whether incumbents (like Tesla with Optimus, or Waymo adapting its tech) or startups will dominate the robotics market.
- Elad Gil argues that capital, hardware, and manufacturing needs favor incumbents, citing Waymo and Tesla as self-driving winners, alongside Chinese car companies.
- Sarah Guo contends that the advantages of existing AI models for robotics are not as strong as perceived, especially for manipulation and hardware, leaving opportunities for startups.
- They discuss the definition of a "robot," agreeing it involves intelligence and generalization beyond single-use appliances like dishwashers.
- Elad Gil emphasizes that self-driving will become truly significant in 2026, impacting personal cars and ride-hailing services.
Future of IPOs and M&A in AI (13:59 - 16:42)
The hosts discuss the financial landscape for AI companies in the coming year:
- Sarah Guo explains that market skittishness could arise from concerns about demand not supporting capital expenditure, systemic risks from credit agreements, or over-concentration in key players like Nvidia.
- She shares an anecdote about a hedge fund manager's strategy to buy AI IPOs regardless of fundamental views, driven by retail appetite and benchmarking pressures.
- Elad Gil predicts a significant increase in AI IPOs in 2026, especially if a major AI company goes public successfully, attracting retail investors and encouraging others to follow suit.
- He notes that IPOs offer a crucial way for AI labs to raise substantial capital.
Challenges in Consumer AI Innovation (16:42 - 21:08)
Sarah Guo expresses optimism for consumer AI, despite past disappointments, and the hosts explore reasons for the slower-than-expected innovation:
- Sarah Guo anticipates a slate of consumer hardware that might mostly fail but is excited about "magical experiences" from new consumer agent software.
- Elad Gil agrees, noting the challenge for startups to achieve "escape velocity" against incumbents who can copy successful features.
- He recalls a Stanford program he ran years ago to encourage consumer AI app development, surprised by the subsequent lack of innovation.
- Reasons for slow consumer AI innovation include incumbents being "scary" and integrating new ideas, founders building "better versions of last generation experiences" rather than truly novel ones, and a limited pool of product people with the necessary creativity and context.
Funding of Neo Labs, RL Research (21:08 - 26:28)
The discussion delves into new research directions and the funding of "Neolabs":
- Elad Gil highlights the surprising funding of new research labs (Neolabs) and the potential of alternative architectures and research in reinforcement learning and continual learning.
- He questions whether these diverse approaches will eventually consolidate into a few dominant players due to the importance of scale and revenue generation.
- Sarah Guo references Ilia Sutskever's description of the "age of research," where compute efficiency and novel ideas can challenge the pure scaling approach.
- She also points to the potential relevance of multiple architectures (diffusion, SSMs) that haven't been scaled yet, and the resource allocation challenges faced by labs balancing inference (revenue) with research.
- Elad Gil speculates on evolutionary systems for AI development, drawing parallels to biology (brain specialization, protein design) where self-evolution and code could lead to rapid advancements.
Predictions for 2026 Beyond AI (26:28 - 30:37)
The hosts offer predictions outside the realm of artificial intelligence:
- Elad Gil predicts an acceleration in defense tech startups and a significant shift towards autonomous and drone-based systems, fundamentally reshaping warfare.
- He notes this trend is driven by current administrations and a growing density of defense tech startups.
- Sarah Guo agrees, acknowledging budget challenges but emphasizing the undeniable need for competitive autonomy in defense, which attracts capital and talent.
- Sarah Guo's non-AI prediction focuses on GLP-1 drugs, stating their impact is still underrated despite widespread enthusiasm.
- She believes their continued adoption will pave the way for other peptide and hormone therapies, with significant second-order effects on public health and willingness to explore engineered peptides.
- Elad Gil adds that the biohacking community, often early adopters of such therapies, serves as an indicator for broader societal adoption of peptides and other longevity/neuromodulation techniques.
2026 Prediction from AI Industry Leaders (30:37 - 40:46)
A segment features predictions from various tech and AI leaders for 2026:
- Jensen Huang (NVIDIA): Reasoning systems will make AI more versatile and robust, revolutionizing every industry from biology to self-driving cars.
- Arvind Jain (Glean): AI will become proactive and deeply integrated into work life, acting as a coach and manager, completing tasks before being asked.
- Winston Weinberg (Harvey): Context will be the most important part of every product, with AI systems focusing on extracting user intent rather than users providing all context.
- Raiza Martin (Huxe): A new suite of product experiences will emerge, running on much faster inference.
- Zach Ziegler (Open Evidence): People will stop copy-pasting into chat boxes; applications will use screen sharing and better context management across sources.
- Scott Wu (Cognition): A mass-scale consumer agentic AI will be created, offering a step function in experience similar to ChatGPT's initial release, with opportunities for startups.
- Aaron Levie (Box): 2026 will be the year of AI agents in the enterprise, particularly in deep vertical or domain-specific areas, requiring workflow integration, data access, context engineering, and change management. He also predicts the rise of "agent harnesses" and economically useful evaluations for knowledge worker tasks.
- Misha Laskin (ReflectionAI): The US will regain leadership in open models, shifting from China, with new labs producing very interesting small open models.
- Noam Brown (OpenAI): AI will become much more politicized, a major discussion point for the 2026 midterm elections.
- Joshua Meier (Chai Discovery): 2025 was the year of research in AI drug discovery, and 2026 will be the year of deployment, with models becoming highly useful for designing complex molecules and antibodies.
- Bryan Johnson (Don't Die): 2026 will be the year "YOLO" (You Only Live Once) dies, replaced by a "don't die" philosophy, driven by a shift from self-destructive behaviors to valuing and defending life, potentially in response to AI's progress.
- Sholto Douglas (Anthropic): Other forms of knowledge work will experience the same transformation as software engineers, with AI doing most of the work. He also predicts continual learning will be solved, first test deployments of home robots, and software engineering going "utterly wild."
- Ben & Asher Spector (Stanford PhDs): 2026 will be the year of energy-efficient AI, as data center power constraints grow. They emphasize the importance of "intelligence per watt" in the short term, but chips matter more in the long term due to faster depreciation.
- Everyone's perceptions about AI will be flipped; the belief that only Nvidia can be used outside Google will be disproven, negative sentiment towards AI will increase, but its perceived usefulness will also flip, driven by transformative utility.
The hosts conclude the episode, thanking listeners and wishing them happy holidays and a happy 2026.
