Mar 7, 2024 · 42m · no-priors
No Priors Ep. 54 | With Sarah Guo & Elad Gil
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 episode of No Priors, Sarah Guo and Elad Gil analyze frontier foundation model developments, the defensibility of AI hardware infrastructure, evolving agent architectures, and the massive economic value being unlocked by enterprise service automation.
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 99.9% of the talking time here. How this is scored →
The hosts as informed peer 7.4 Guest teaching 2.1 Guest disagreement 1.4 The hosts pushing back 1.4
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 38:06 Sarah challenges domestic fab viability
Sarah counters Elad's argument by emphasizing that TSMC's dominance relies heavily on human capital and corporate culture that cannot easily be replicated domestically.
Hardest push from the hosts ▶ 37:51 Elad rejects domestic manufacturing absence premiseElad firmly pushes back against the notion that the US lacks manufacturing human capital by citing Intel and Texas Instruments' multi-decade domestic production track records.
Biggest teaching moment ▶ 32:01 Sarah details manufacturing and yield moatsSarah expands the hardware moat beyond silicon design and interconnect to include TSMC allocation constraints, packaging yield, and emerging state space model architectures.
The host holds their own ▶ 24:05 Elad models services spend conversion to softwareElad delivers precise quantitative modeling comparing $500B annual software spend to $3.5T-$5T in addressable services payroll to demonstrate generative AI's market potential.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Foundation Model Developments and Context Windows | 7 | 2 | 1 | 1 | The co-hosts explore recent model announcements collaboratively. Elad provides domain depth on biological context windows and protein sizes, while Sarah analyzes Mistral's strategic focus on efficiency and the RAG versus large context tradeoff. | |
| Inference-Time Compute and Google's AI Position | 8 | 2 | 2 | 2 | Elad draws analogies between inference-time compute in game AI and future agentic reasoning. They engage in a nuanced discussion evaluating Google's internal will, distribution advantages, and recent Gemini release velocity. | |
| Specialized AI Domains and Systemic Agent Architectures | 7 | 2 | 1 | 1 | Both hosts analyze specialized domain models in biology and robotics. Sarah outlines the shift from open-ended brittle agents toward constrained systems with verifiable reward feedback loops, which Elad reinforces. | |
| NVIDIA Earnings, AI CapEx, and Big Tech Value Accrual | 8 | 2 | 1 | 1 | Sarah details portfolio insights on GPU cluster upgrade economics and Meta's CapEx-to-ROI conversion. Elad calculates Azure's annualized AI run rate and contextualizes big tech's outsized value capture relative to venture-backed startups. | |
| Enterprise Adoption, Service Automation, and the Application Wave | 8 | 2 | 1 | 1 | Elad cites granular operational statistics from Klarna's customer service bot replacing 700 workers and quantifies the multi-trillion-dollar addressable payroll market. Sarah maps out the commercial adoption dynamics across tech incumbents and SMB platforms. | |
| Hardware Moats, Semiconductor Manufacturing, and Geopolitics | 8 | 4 | 3 | 3 | The conversation turns into an active debate regarding semiconductor moats and US domestic fab viability. Sarah points out TSMC's cultural and yield barriers, while Elad pushes back citing Intel's decades-long domestic manufacturing history before Sarah notes Intel's lag in process node technology. | |
| Feedback Loops, Rapid Innovation Cycles, and RLPAF | 6 | 1 | 1 | 1 | The hosts conclude with a high-level review of the compounding virtuous cycle in AI funding, corporate adoption surveys, and coin a playful acronym for adoption feedback loops. |
Statements from this episode (21)
Gil: Longer context windows will prove more important than expected for biology
“I think for certain application areas, like biology, longer context windows actually seem to be quite important. And so, for example, if you're doing a protein folding model and you have a short context window, you're often actually not encapsulating much of t…”
Gil: Mistral reached near GPT-4 capability within one year of founding
“They went from basically starting the company to almost GPT-IV level in less than a year.”
Guo: Large context windows expand RAG trade-offs rather than killing retrieval
“I'm more of the belief that it just opens up the set of trade-offs you can make between retrieval, more sophisticated retrieval and model reasoning by having a larger context window versus saying, like, we don't need any ability to work with a specific data se…”
Gil: Future AI reasoning will shift to inference-time compute and retraining
“And I think that's also a little bit under discussed in terms of probably a lot of what's going to happen in the future, particularly we get into agents and reasoning is stuff that's happening at that point of inference. And then it's used to sort of feedback …”
Guo: Google's unreleased internal multimodal and function-calling capabilities are highly impressive
“Gemini is a very impressive model. I think the capabilities that they have internally that they haven't released yet around additional, like, function calling and multimodality are also really, really impressive”
Gil: Competitive pressure has reignited Google's will to make major AI strides
“And so really, I think the thing that was lacking until recently was the will. And it seems like now, because of the competitive dynamic, the will has been reborn, right? And so it really feels to me like they are gonna make really big strides going forward.”
Guo: ChatGPT cannot design a DNA sequence expressing CRISPR-Cas9
“As a task, for example, if you ask ChatGPT to design a DNA sequence that can express Prisper Cas- nine, it can't do that yet, right?”
Gil: Specialized AI models will proliferate across hard sciences in 2024-2025
“So it seems like the, you know, in general, I wouldn't be surprised if 20, 24 and 20, 25 is the year of proliferation of models where we're going to start to see an expansion in terms of the different types that are covered, you know, chemistry and material sc…”
Gil: New reinforcement learning AI agent products will emerge within 6-12 months
“And so I think that that purpose of knowledge is about to hit the world in the context of new products. And it'll take time for those products to emerge, you know, six months, 12 months, a year. But It does feel like that's another wave that's coming where you…”
Guo: Portfolio company is buying tens of thousands of unreleased B100 GPUs
“I was talking to one of my portfolio companies that's buying in the tens of thousands of GPU size and is skipping to B 100 because they described it as like free money in terms of training efficiency.”
Gil: Big Tech, not VCs, drives multi-billion dollar AI funding rounds
“The big rounds aren't venture capitalists. Investing billions of dollars is the big tech companies. It's Amazon and Google and Microsoft and Salesforce and NVIDIA actually, right?”
Guo: Meta Added $197B in Largest Single-Session Gain Pre-NVIDIA
“They had this one day ad of a hundred and ninety seven billion of market cap. Biggest single session ad before NVIDIA.”
Gil: Incumbent tech companies will capture most near-term AI market value
“But at least for the next few years, it seems like where we're going to see that Really huge market cap incremental add maybe companies like OpenAI and some of the model companies, but also it seems like increasingly it's just going to be existing companies ad…”
Guo: Most legacy services companies will fail to make AI transition
“I think the answer is mostly, especially some of these services firms, like maybe they partner to get there, but they mostly will not make the transition, I think.”
Gil: Klarna's AI assistant performed the equivalent work of 700 human agents
“One of the folks from Klarna posted today that they built an AI assistant that's powered by OpenAI that in its first four weeks handled 2.3 million customer service chats for them. And so it ended up handling two thirds of all their customer service inquiries.…”
Gil: Converting 10% of human services payroll to AI matches US software
“If you look at spend on software in the U S right now, it's about half trillion dollars. And software spend a year, if you look at, ah, human-centric services, just payroll, for things where gen AI can probably impact things, it's three and a half to five tril…”
Gil: Advanced semiconductor markets historically produce only one leader and one runner-up
“If you look at many of the most advanced chip markets, historically at least, there's tended to be a leader, and then there's been, there's tended to be a second place party, and that was
You know, during the microprocessor world, that was Intel, and then AMD …”
Guo: NVIDIA's competitive moat is really deep despite intense economic pressure
“I think the desire, like the economic pressure given two trillion of market cap and
More demand than NVIDIA can support is higher than ever, but I think the moat is actually really, really deep.”
Guo: Has not yet found a compelling AI chip startup to fund
“I would love to meet companies in this area and still haven't, haven't seen something that's gotten me over the edge, even in a place that it's so
Obviously, economically fertile.”
Gil: Japan may emerge as a key semiconductor manufacturing hedge against Taiwan
“And so I'm increasingly wondering whether Japan emerges as sort of a second source location and part to geopolitically hedge Taiwan.”
Gil: Rapid AI Breakthroughs Continually Invalidate Static Hypotheses
“The more I learn, the less I know in AI, and it's the opposite of every other field I've ever been in. Usually the more you learn about something, yeah, usually the more you learn about something, the more you can create sort of these straight line hypotheses,…”