Dec 26, 2024 · 27m · no-priors

No Priors Ep. 95 | Best of 2024

Sarah Guo · 5m spoken Bret Taylor · 3m spoken Jensen Huang · 2m spoken Dmitri Dolgov · 2m spoken Alexandr Wang · 2m spoken Andrej Karpathy · 1m spoken Elad Gil · 1m spoken
0:00 / 0:00
▶ Watch on YouTube →

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

In this 2024 retrospective episode of No Priors, hosts Sarah Guo and Elad Gil highlight foundational conversations with AI leaders including Jensen Huang, Andrej Karpathy, Bret Taylor, and Alexandr Wang, exploring data center compute, agentic interfaces, world modeling, and the incremental path to AGI.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 30.1% of the talking time here. How this is scored →

The hosts as informed peer 5.6 Guest teaching 5.9 Guest disagreement 2.0 The hosts pushing back 2.3
05100:0010:0020:000:05–3:59 · The hosts as informed peer 5/10 Jensen Huang on NVIDIA and Data Center Computing Sarah sets up the discussion on NVIDIA's expansion from chips to racks and data centers. Jensen elaborates on why software requires building full data centers to test real-world fabrics rather than relying on PowerPoint slides.3:59–7:13 · The hosts as informed peer 6/10 Andrej Karpathy on AI Exocortex and Model Distillation Andrej argues that renting an exocortex feels invasive and draws an analogy to crypto keys, while Sarah pushes back by noting users will happily sacrifice ownership to rent a smarter brain.7:13–11:17 · The hosts as informed peer 5/10 Bret Taylor on Branded AI Company Agents Bret explains branded enterprise agents using the 1995 website metaphor. Elad briefly interjects to anchor the theoretical roadmap into concrete customer products like Sonos and SiriusXM.11:17–15:52 · The hosts as informed peer 6/10 OpenAI Sora Team on World Models and Visual AI Sarah questions whether Sora mirrors approximate human mental physics or acts like a rigorous physics engine. Bill explains that human approximation is an evolutionary limitation and raw data prediction scaling will exceed human fidelity.15:53–19:00 · The hosts as informed peer 4/10 Dmitry Dolgov on the Long Tail of Autonomous Driving Sarah asks about the distinction between driver-assist and full autonomy. Dmitry educates on the deceptive ease of building initial prototypes versus solving the long tail of reliability nines.19:00–23:28 · The hosts as informed peer 6/10 Dylan Field on the Evolution of AI Interfaces Elad asks if multimodal AI will make voice or agentic interfaces eliminate graphical UIs. Dylan pushes back against the hype, arguing new modalities complement rather than replace traditional software interfaces.23:29–26:28 · The hosts as informed peer 7/10 Alexandr Wang on the Incremental Road to AGI Alex contrarianly compares AGI progress to curing cancer rather than discovering a single vaccine. Sarah demonstrates strong domain depth by asking whether the lack of generality stems from MCTS limits or scaling walls, prompting Alex to reject cross-modal transfer narratives.0:05–3:59 · Guest teaching 6/10 Jensen Huang on NVIDIA and Data Center Computing Sarah sets up the discussion on NVIDIA's expansion from chips to racks and data centers. Jensen elaborates on why software requires building full data centers to test real-world fabrics rather than relying on PowerPoint slides.3:59–7:13 · Guest teaching 6/10 Andrej Karpathy on AI Exocortex and Model Distillation Andrej argues that renting an exocortex feels invasive and draws an analogy to crypto keys, while Sarah pushes back by noting users will happily sacrifice ownership to rent a smarter brain.7:13–11:17 · Guest teaching 5/10 Bret Taylor on Branded AI Company Agents Bret explains branded enterprise agents using the 1995 website metaphor. Elad briefly interjects to anchor the theoretical roadmap into concrete customer products like Sonos and SiriusXM.11:17–15:52 · Guest teaching 6/10 OpenAI Sora Team on World Models and Visual AI Sarah questions whether Sora mirrors approximate human mental physics or acts like a rigorous physics engine. Bill explains that human approximation is an evolutionary limitation and raw data prediction scaling will exceed human fidelity.15:53–19:00 · Guest teaching 7/10 Dmitry Dolgov on the Long Tail of Autonomous Driving Sarah asks about the distinction between driver-assist and full autonomy. Dmitry educates on the deceptive ease of building initial prototypes versus solving the long tail of reliability nines.19:00–23:28 · Guest teaching 5/10 Dylan Field on the Evolution of AI Interfaces Elad asks if multimodal AI will make voice or agentic interfaces eliminate graphical UIs. Dylan pushes back against the hype, arguing new modalities complement rather than replace traditional software interfaces.23:29–26:28 · Guest teaching 6/10 Alexandr Wang on the Incremental Road to AGI Alex contrarianly compares AGI progress to curing cancer rather than discovering a single vaccine. Sarah demonstrates strong domain depth by asking whether the lack of generality stems from MCTS limits or scaling walls, prompting Alex to reject cross-modal transfer narratives.0:05–3:59 · Guest disagreement 2/10 Jensen Huang on NVIDIA and Data Center Computing Sarah sets up the discussion on NVIDIA's expansion from chips to racks and data centers. Jensen elaborates on why software requires building full data centers to test real-world fabrics rather than relying on PowerPoint slides.3:59–7:13 · Guest disagreement 2/10 Andrej Karpathy on AI Exocortex and Model Distillation Andrej argues that renting an exocortex feels invasive and draws an analogy to crypto keys, while Sarah pushes back by noting users will happily sacrifice ownership to rent a smarter brain.7:13–11:17 · Guest disagreement 1/10 Bret Taylor on Branded AI Company Agents Bret explains branded enterprise agents using the 1995 website metaphor. Elad briefly interjects to anchor the theoretical roadmap into concrete customer products like Sonos and SiriusXM.11:17–15:52 · Guest disagreement 1/10 OpenAI Sora Team on World Models and Visual AI Sarah questions whether Sora mirrors approximate human mental physics or acts like a rigorous physics engine. Bill explains that human approximation is an evolutionary limitation and raw data prediction scaling will exceed human fidelity.15:53–19:00 · Guest disagreement 1/10 Dmitry Dolgov on the Long Tail of Autonomous Driving Sarah asks about the distinction between driver-assist and full autonomy. Dmitry educates on the deceptive ease of building initial prototypes versus solving the long tail of reliability nines.19:00–23:28 · Guest disagreement 3/10 Dylan Field on the Evolution of AI Interfaces Elad asks if multimodal AI will make voice or agentic interfaces eliminate graphical UIs. Dylan pushes back against the hype, arguing new modalities complement rather than replace traditional software interfaces.23:29–26:28 · Guest disagreement 4/10 Alexandr Wang on the Incremental Road to AGI Alex contrarianly compares AGI progress to curing cancer rather than discovering a single vaccine. Sarah demonstrates strong domain depth by asking whether the lack of generality stems from MCTS limits or scaling walls, prompting Alex to reject cross-modal transfer narratives.0:05–3:59 · The hosts pushing back 1/10 Jensen Huang on NVIDIA and Data Center Computing Sarah sets up the discussion on NVIDIA's expansion from chips to racks and data centers. Jensen elaborates on why software requires building full data centers to test real-world fabrics rather than relying on PowerPoint slides.3:59–7:13 · The hosts pushing back 3/10 Andrej Karpathy on AI Exocortex and Model Distillation Andrej argues that renting an exocortex feels invasive and draws an analogy to crypto keys, while Sarah pushes back by noting users will happily sacrifice ownership to rent a smarter brain.7:13–11:17 · The hosts pushing back 2/10 Bret Taylor on Branded AI Company Agents Bret explains branded enterprise agents using the 1995 website metaphor. Elad briefly interjects to anchor the theoretical roadmap into concrete customer products like Sonos and SiriusXM.11:17–15:52 · The hosts pushing back 3/10 OpenAI Sora Team on World Models and Visual AI Sarah questions whether Sora mirrors approximate human mental physics or acts like a rigorous physics engine. Bill explains that human approximation is an evolutionary limitation and raw data prediction scaling will exceed human fidelity.15:53–19:00 · The hosts pushing back 1/10 Dmitry Dolgov on the Long Tail of Autonomous Driving Sarah asks about the distinction between driver-assist and full autonomy. Dmitry educates on the deceptive ease of building initial prototypes versus solving the long tail of reliability nines.19:00–23:28 · The hosts pushing back 2/10 Dylan Field on the Evolution of AI Interfaces Elad asks if multimodal AI will make voice or agentic interfaces eliminate graphical UIs. Dylan pushes back against the hype, arguing new modalities complement rather than replace traditional software interfaces.23:29–26:28 · The hosts pushing back 4/10 Alexandr Wang on the Incremental Road to AGI Alex contrarianly compares AGI progress to curing cancer rather than discovering a single vaccine. Sarah demonstrates strong domain depth by asking whether the lack of generality stems from MCTS limits or scaling walls, prompting Alex to reject cross-modal transfer narratives.

speaking balance: gold is the hosts, purple is the guest (3 minute bins)

0:00 · the hosts 29.6% · guest 70.4%0:00 · the hosts 29.6% · guest 70.4%3:00 · the hosts 32.9% · guest 67.1%3:00 · the hosts 32.9% · guest 67.1%6:00 · the hosts 20.3% · guest 79.7%6:00 · the hosts 20.3% · guest 79.7%9:00 · the hosts 27.6% · guest 72.4%9:00 · the hosts 27.6% · guest 72.4%12:00 · the hosts 25.2% · guest 74.8%12:00 · the hosts 25.2% · guest 74.8%15:00 · the hosts 24.7% · guest 75.3%15:00 · the hosts 24.7% · guest 75.3%18:00 · the hosts 36.8% · guest 63.2%18:00 · the hosts 36.8% · guest 63.2%21:00 · the hosts 41.2% · guest 58.8%21:00 · the hosts 41.2% · guest 58.8%24:00 · the hosts 31% · guest 69%24:00 · the hosts 31% · guest 69%27:00 · the hosts 100% · guest 0%27:00 · the hosts 100% · guest 0%
Sharpest disagreement ▶ 26:05 Alexandr Wang rejects video world model narrative

Alex flatly dismisses the popular industry narrative that video training transfers to broader intelligence, stating there is no scientific evidence of cross-modal transfer.

Hardest push from the hosts ▶ 5:20 Sarah Guo rejects the exocortex ownership premise

Sarah directly counters Andrej Karpathy's insistence on owning weights by stating she would willingly surrender ownership to rent a superior brain.

Biggest teaching moment ▶ 16:31 Dmitry Dolgov on the illusion of autonomous driving progress

Dmitry illustrates that Waymo achieved a 100-mile zero-intervention run back in 2009 with tiny compute, demonstrating that prototype success obscures the brutal reality of solving the long tail of safety nines.

The host holds their own ▶ 25:12 Sarah Guo dissects technical barriers to AGI generality

Sarah demonstrates deep technical command by framing Alex's high-level analogy into specific research bottlenecks like Monte Carlo Tree Search limitations and multi-step reasoning walls.

the scores for every segment, with the reasoning behind each
ChapterTopicThe hosts as informed peerGuest teachingGuest disagreementThe hosts pushing backWhy
Jensen Huang on NVIDIA and Data Center Computing 5621 Sarah sets up the discussion on NVIDIA's expansion from chips to racks and data centers. Jensen elaborates on why software requires building full data centers to test real-world fabrics rather than relying on PowerPoint slides.
Andrej Karpathy on AI Exocortex and Model Distillation 6623 Andrej argues that renting an exocortex feels invasive and draws an analogy to crypto keys, while Sarah pushes back by noting users will happily sacrifice ownership to rent a smarter brain.
Bret Taylor on Branded AI Company Agents 5512 Bret explains branded enterprise agents using the 1995 website metaphor. Elad briefly interjects to anchor the theoretical roadmap into concrete customer products like Sonos and SiriusXM.
OpenAI Sora Team on World Models and Visual AI 6613 Sarah questions whether Sora mirrors approximate human mental physics or acts like a rigorous physics engine. Bill explains that human approximation is an evolutionary limitation and raw data prediction scaling will exceed human fidelity.
Dmitry Dolgov on the Long Tail of Autonomous Driving 4711 Sarah asks about the distinction between driver-assist and full autonomy. Dmitry educates on the deceptive ease of building initial prototypes versus solving the long tail of reliability nines.
Dylan Field on the Evolution of AI Interfaces 6532 Elad asks if multimodal AI will make voice or agentic interfaces eliminate graphical UIs. Dylan pushes back against the hype, arguing new modalities complement rather than replace traditional software interfaces.
Alexandr Wang on the Incremental Road to AGI 7644 Alex contrarianly compares AGI progress to curing cancer rather than discovering a single vaccine. Sarah demonstrates strong domain depth by asking whether the lack of generality stems from MCTS limits or scaling walls, prompting Alex to reject cross-modal transfer narratives.

Statements from this episode (13)

Insight
Huang: The data center is the new unit of computing
“You know, I say that the new unit of computing is the data center.”
Jensen Huang Dec 26, 2024 ▶ 1:50
Disclosure
Huang: NVIDIA will build five more internal supercomputers next year
“We have five supercomputers in our company today. Next year, we're going to build easily five more.”
Jensen Huang Dec 26, 2024 ▶ 2:11
Insight
Huang: Companies serious about software must build their own computers
“So if you're serious about software, you build your own computers. If you're serious about software, then you're going to build your whole computer, and we build it all at scale.”
Jensen Huang Dec 26, 2024 ▶ 2:16
Disclosure
Huang: NVIDIA optimizes full-stack infrastructure then sells disaggregated parts
“We build it at scale, and we build it very vertically integrated. We optimize it full stack, and then we disaggregate everything, and we sell it in parts.”
Jensen Huang Dec 26, 2024 ▶ 2:25
Insight
Karpathy: LLMs waste massive capacity memorizing irrelevant data like hashes
“And I do think that the current models are wasting a ton of capacity remembering stuff that doesn't matter. Like they remember SHA hashes.”
Andrej Karpathy Dec 26, 2024 ▶ 6:29
Prediction Not checkable as stated
Karpathy: Cognitive core models will likely shrink to one billion parameters
“I think even a billion. Billions of ISIS. We'll probably get to that point and the models can be very, very small.”
Andrej Karpathy Dec 26, 2024 ▶ 6:54
Prediction Not checkable as stated
Taylor: In 2025, digital presence probably means having branded AI agents
“In 2025, existing digitally will probably mean having a branded AI agent that your customers can interact with to do everything that they can do on your website.”
Bret Taylor Dec 26, 2024 ▶ 7:49
Insight
Dolgov: AV complexity lies in long-tail reliability, not prototypes or ADAS
“The hard problem, and it's kind of like that, the early part of the curve has been getting like, you know, even steeper and steeper, but that's not where the complexity is. The complexity is in the long tail of the many, many, many nines. And you don't see tha…”
Dmitri Dolgov Dec 26, 2024 ▶ 17:22
Prediction Not checkable as stated
Field: AI agents will lead to more user interfaces and software
“And I think that the interfaces will get more sophisticated. But also even if they don't, I suspect that it's just like any new media type when it's introduced, it's not like the old media types go away, right? Just because you have TikTok doesn't mean that yo…”
Dylan Field Dec 26, 2024 ▶ 20:00
Prediction Not checkable as stated
Wang: The path to AGI resembles curing cancer, not a vaccine
“My biggest belief here is that the path to AGI is is one that looks a lot more like curing cancer than developing a vaccine. And what I mean by that is I think that the path to build AGI is going to be in, in, you know, you're going to have to solve a bunch of…”
Alexandr Wang Dec 26, 2024 ▶ 23:49
Prediction Not checkable as stated
Wang: Society will adapt smoothly due to slow AI progress
“I think it's actually a pretty bullish case for society adapting the technology because I think it's going to be, you know, consistent, slow progress. For quite some time, and society will have time to fully sort of acclimate to the technology that develops.”
Alexandr Wang Dec 26, 2024 ▶ 24:59
Assertion Contradicted
Wang: AI gets no positive transfer across modalities like video to text
“My understanding there's no positive transfer from learning in one modality to other modalities. So like training off of a bunch of video doesn't really help you that much with your text problems and vice versa.”
Alexandr Wang Dec 26, 2024 ▶ 25:42
Assertion Not checkable as stated
Wang: No strong evidence that video creates useful AI world models
“I don't think there's strong scientific evidence of that yet. Maybe there will be eventually.”
Alexandr Wang Dec 26, 2024 ▶ 26:11
Made with StarZero

Turn any episode into a week of clips.

This entire site, over 100 episodes transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.