Jan 11, 2024 · 19m · no-priors
No Priors Ep. 46 | Best of 2023 with Sarah Guo and Elad Gil
Mustafa Suleyman · 2m spoken Reid Hoffman · 2m spoken Sarah Guo · 2m spoken Arthur Mensch · 2m spoken Ilya Sutskever · 2m spoken Alyssa Henry · 1m spoken Daphne Koller · 1m spoken Jensen Huang · 1m spoken Noam Shazeer · 56s 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
This retrospective episode of No Priors compiles highlights from 2023 interviews with top artificial intelligence leaders, exploring frontier technical architectures, labor economics, safety governance, and the future trajectory of AI innovation.
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 13.7% of the talking time here. How this is scored →
The hosts as informed peer 0.7 Guest teaching 1.7 Guest disagreement 1.0 The hosts pushing back 0.2
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 15:35 Rejecting built-in safety assumptions
Arthur Mensch directly counters prevailing industry consensus by arguing that expecting raw base models to be well-behaved is a flawed premise.
Hardest push from the hosts ▶ 6:45 Sarah tests Suleiman's core definitionSarah directly interrupts Suleiman's historical overview to challenge whether he still subscribes to the early DeepMind formulation of intelligence.
Biggest teaching moment ▶ 6:50 Mustafa critiques industry over-rotationMustafa Suleiman systematically breaks down why OpenAI and Anthropic's pursuit of monolithic generality misses the importance of contextual attention and dynamic routing.
The host holds their own ▶ 11:40 Sarah contextualizes graphical models in biotechSarah succinctly connects abstract foundational architectures like probabilistic graphical models to practical economic breakthroughs in drug discovery.
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 |
|---|---|---|---|---|---|---|
| Alyssa Henry on AI Unlocking Small Business Productivity | 0 | 1 | 0 | 0 | Sarah Guo introduces an archival clip where Alyssa Henry discusses how generative AI tools solve the marketing pain points and unaddressed demand for small business owners. Because this is a curated clip reel, there is no direct host-guest interaction or pushback. | |
| Mustafa Suleiman on Defining Intelligence and System Routers | 2 | 3 | 2 | 1 | Mustafa Suleiman explains Shane Legg's early definition of AGI before Sarah asks if that remains his working definition today. Suleiman reframes the premise, arguing the industry has over-rotated on generality when contextual routing across specialized tools is the true breakthrough. | |
| Reid Hoffman on Human-AI Symbiosis and Technological Disruption | 0 | 1 | 1 | 0 | Reid Hoffman provides an uninterrupted monologue framing AI as a 'steam engine of the mind' and pushing back against historical doomsaying cycles like those seen with the printing press. The host provides only the lead-in context. | |
| Daphne Koller on Deep Learning and Causal Reasoning in Biotech | 2 | 2 | 1 | 0 | Sarah introduces snippets from Daphne Koller on probabilistic graphical models versus deep learning pattern recognition, followed by Noam Shazeer defending why text density matters more than other modalities. | |
| Arthur Mensch on Open AI Models and Modular Guardrails | 0 | 2 | 2 | 0 | Arthur Mensch presents Mistral's design philosophy, explicitly rejecting the assumption that raw foundational models should be intrinsically constrained rather than managed via modular external filters and market competition. | |
| Jensen Huang on NVIDIA's Dual Innovation Architecture | 0 | 1 | 0 | 0 | Jensen Huang describes NVIDIA's dual organizational structure balancing disciplined execution on supercomputing with agile, 10-year horizon skunkworks exploration before Sarah closes the retrospective episode. |
Statements from this episode (8)
Suleiman: OpenAI and Anthropic over-rotate on single-agent generality
“I think that's a good definition of intelligence, but I think in a weird way, it's over-rotated the entire field on one aspect of intelligence, which is generality, you know, and I think OpenAI and then subsequently Anthropic and others have taken up this defa…”
Suleiman: Central system routers will unlock AI rather than giant models
“And in fact, I think the key, and we can get into this obviously in a moment, but I think the key element that is going to really unlock this field is actually going to be the router. In the middle of a series of different systems which are specialized, some o…”
Koller: AI requires synthesizing deep learning with causal and interpretable modeling
“What I think we're starting to see right now is a the pendulum starting to swing back in the sense that there is a greater understanding that you really need a bit of both. You need that hugely powerful pattern recognition that we get from deep learning, but y…”
Shazeer: Text Is a Thousand Times More Information-Dense Than Images
“An image is worth a thousand words, but it's like a million pixels. So like the text is like a thousand times as dense.”
Shazeer: Most Core AI Intelligence Will Come From Text Models
“But I think that a lot of the core intelligence is going to come from a From these text models.”
Mensch: AI base models should know everything and use modular guardrails
“Assuming that the model should be well behaved is, I think, a wrong assumption. You need to make the assumption that the model should know everything. And then on top of that, have some modules that moderate and guardrail the model.”
Mensch: AI safety should regulate application makers, not trust big model labs
“And the way you make this healthy competition is not By trusting a couple of companies to do their own safety, it's rather for, it's rather the way you do it is to ask application makers to comply with some rules.”
Huang: NVIDIA Operates Two Parallel Systems for Long-Term Innovation and Perfect Execution
“Our company actually has two different ways of working. One of them is rather organic, shape-shifting all the time. If a particular investment is not working out, we give up on it, move the resources somewhere else. And so that's the agile part of the company.…”