Explore intelligence

Report · Money

Examples and radical ideas: Shenzhen, Prospera, seasteading, Mars, Freedom Cities

Armstrong cites Shenzhen as existing as a large-scale manufacturing region in part because it became a special economic zone, and describes Prospera in Honduras, in which Coinbase invested just to learn more about it; Prospera has had a mixed up-and-down relationship with the local Honduran government but is getting closer to a resolution, in his understanding. He mentions radical solutions discussed like seasteading, Mars, and brain uploading to cyberspace, and the Freedom Cities idea from the campaign: taking 10 or 100 square miles of US federal land, exempting it from normal red tape (EPA, local community approval) to allow building data centers, nuclear power plants, and unlimited drone delivery without FAA approval. He notes the federal government owns many pockets of land, that zones

No Priors: Artificial Intelligence | Technology | Startups · Coinbase’s Everything Exchange: Agentic Finance, Stablecoins, and Tokenization with CEO Brian Armstrong

Claim from Brian Armstrong

Prospera status is Armstrong's understanding, not verified; Freedom Cities is a campaign idea.

Prediction · Money

Embryo editing will become normal; eventually not doing it will seem risky

Armstrong predicts embryo editing, though controversial today and familiar from movies like Gattaca, will become more normal over time, and that at some point it will be abnormal to have children without this kind of screening and editing, because not doing it would be considered risky. He grounds this in the observation that nobody wants rare genetic diseases for their children and everybody wants healthier, better lives for their children.

No Priors: Artificial Intelligence | Technology | Startups · Coinbase’s Everything Exchange: Agentic Finance, Stablecoins, and Tokenization with CEO Brian Armstrong

Claim from Brian Armstrong

Predictions hedged with 'I think' and 'at some point'.

Report · Money

Card networks do not work well for transactions under $1

Arguably, card networks do not work well for anything under $1 — the fee economics relative to transaction size break down at very small amounts.

No Priors: Artificial Intelligence | Technology | Startups · Coinbase’s Everything Exchange: Agentic Finance, Stablecoins, and Tokenization with CEO Brian Armstrong

Claim from Brian Armstrong

Report · Money

Epigenetic reprogramming idea traced to early lab results, including Yamanaka's

Armstrong recounts that the epigenetic reprogramming idea came out of expert dinners, and that some labs, like Shinya Yamanaka's, had seen early results showing cells can be reprogrammed to restore function they had when younger. The host adds that Yamanaka factors changed the imprint for both age and cell type.

No Priors: Artificial Intelligence | Technology | Startups · Coinbase’s Everything Exchange: Agentic Finance, Stablecoins, and Tokenization with CEO Brian Armstrong

Claim from Brian Armstrong

Transcript spells 'Amanaka'; likely Yamanaka factors.

Report · Money

Coinbase's internal agent harness Toshi pays vendors via X402

Coinbase's internal agent harness, Toshi, can make calls to external vendors and pay via agentic finance over X402 protocols.

No Priors: Artificial Intelligence | Technology | Startups · Coinbase’s Everything Exchange: Agentic Finance, Stablecoins, and Tokenization with CEO Brian Armstrong

Claim from Brian Armstrong

Report · Money

New Limit's reprogramming progress: mice, primates, phase one next year

New Limit has demonstrated successful reprogramming of at least one human cell type in humanized mouse models, is now testing in non-human primates, and will launch its first phase one clinical trial next year.

No Priors: Artificial Intelligence | Technology | Startups · Coinbase’s Everything Exchange: Agentic Finance, Stablecoins, and Tokenization with CEO Brian Armstrong

Claim from Brian Armstrong

Report · Money

88% of Coinbase revenue now comes from non-Bitcoin trading

Per Coinbase's recent earnings deck, 88% of its revenue now comes from non-Bitcoin trading.

No Priors: Artificial Intelligence | Technology | Startups · Coinbase’s Everything Exchange: Agentic Finance, Stablecoins, and Tokenization with CEO Brian Armstrong

Claim from Brian Armstrong

Creator perspective · Money

Freedom Cities: contained deregulation zones on federal land

Armstrong describes the Freedom Cities idea, discussed at one point during a campaign, which proposed taking 10 to 100 square miles of US federal land and creating zones where experimentation can happen in a safe, contained way to see what works and what doesn't. In this concept, designated zones would allow building data centers, nuclear power plants, and unlimited drone delivery without EPA or FAA approval, as a way to experiment with deregulation.

No Priors: Artificial Intelligence | Technology | Startups · Coinbase’s Everything Exchange: Agentic Finance, Stablecoins, and Tokenization with CEO Brian Armstrong

Claim from Brian Armstrong

moderate

Report · Money

agentic.market lists top service providers for search and financial data

The agentic.market website lists the top service providers doing search and financial data — a directory of the emerging agentic services economy.

No Priors: Artificial Intelligence | Technology | Startups · Coinbase’s Everything Exchange: Agentic Finance, Stablecoins, and Tokenization with CEO Brian Armstrong

Claim from Brian Armstrong

Anecdote · Money

Founding New Limit and Armstrong's investor role

Through a series of dinners, Armstrong found the right group of people who came together to found New Limit: Jacob Kimmel (now CEO), Blake Byers, and Greg Johnson. He put initial money into the company and committed to helping with company building, but the science was totally new to him so he was basically a sponge of knowledge soaking it up. He keeps Coinbase as his full-time job and is basically just an investor and board member at New Limit, spinning up another company and trying to see where he can be helpful while staying humble about the things he doesn't know.

No Priors: Artificial Intelligence | Technology | Startups · Coinbase’s Everything Exchange: Agentic Finance, Stablecoins, and Tokenization with CEO Brian Armstrong

Claim from Brian Armstrong

Anecdote · Money

Former LightMatter engineer on ADC/DAC accuracy problems

A former LightMatter engineer who worked on the DAC and ADC subsystem for a silicon photonics accelerator identified ADC/DAC resolution and accuracy as a problem that eventually affected prediction correctness over time in the field.

Y Combinator · What If We Stopped Using GPUs? | YC Paper Club

Claim from guest (former LightMatter engineer)

anecdotal; speaker cannot discuss the specific project

Prediction · Money

AI agents will likely need a stablecoin-backed credit card for backward compatibility

Armstrong expects AI agents will probably need a stablecoin-backed credit card for backward compatibility with existing payment rails.

No Priors: Artificial Intelligence | Technology | Startups · Coinbase’s Everything Exchange: Agentic Finance, Stablecoins, and Tokenization with CEO Brian Armstrong

Claim from Brian Armstrong

high

Recommendation · Money

Act on persistent ideas quickly — motivation is perishable

Armstrong advises that when an idea won't leave your mind for months — like he couldn't stop thinking about epigenetic reprogramming for three months, the way he felt about Bitcoin in 2010-2011 — treat it as a signal worth acting on, and act on a moment of inspiration quickly because motivation is perishable and will fade.

No Priors: Artificial Intelligence | Technology | Startups · Coinbase’s Everything Exchange: Agentic Finance, Stablecoins, and Tokenization with CEO Brian Armstrong

Recommendation from Brian Armstrong

Report · Money

The culture optimizes for reducing entropy, not feeling good or bad

The neurons themselves don't feel bad; they have no pain receptors and no noxious function or poison is introduced to the cells. The brain culture doesn't optimize for feeling good or feeling bad; it optimizes for reducing the entropy in the culture itself. The culture wants to reduce the rate at which it gets asynchronous stimulation and will self-organize to prevent it. The system never reaches an equilibrium; there are two dynamical systems almost fighting for control, pushing themselves to better and better outputs over time. Entropy is the enemy: it disrupts the two dynamical systems, pushing them towards semi-attractor states.

Y Combinator · What If We Stopped Using GPUs? | YC Paper Club

Claim from Y Combinator

Interpretive claims by the speaker about what the culture optimizes.

Creator perspective · Money

Brain as input-output machine framing

The approach treats the brain as an input-output machine where a set of stimuli produces intelligence that yields computationally useful output.

Y Combinator · What If We Stopped Using GPUs? | YC Paper Club

Claim from unknown

framing, not a proven fact

Report · Money

Team actively penalized entropy to encourage exploration

The team actively penalized entropy because they found it helpful, and the cells provided sufficient entropy for exploration on their own.

Y Combinator · What If We Stopped Using GPUs? | YC Paper Club

Claim from unknown

Creator perspective · Money

Optics is probably the right answer for compute; data centers waste energy converting light to electrons

The host argues optics is probably the right answer for compute: data centers already communicate in light, then wastefully switch to electrons and back. The signal path goes light → switch → electrons → switch → light → DAC → light → ADC → light, burning 99.99% of the joules in the conversions. Doing the whole computation in light would be instantaneous and draw no watts — which he thinks the aliens are probably doing. He remains open to other ideas including neural computing and neuromorphics.

Y Combinator · What If We Stopped Using GPUs? | YC Paper Club

Claim from Y Combinator host

Creator perspective · Money

Why optical compute is efficient: passive weight modulation, parallelism, and N-vs-N² energy scaling

Optics is by definition massively parallelizable because photons don't interact with other photons — many optical beams can coexist in a given space without affecting each other, whereas in electrical circuits you have to define different paths for each pair of signals. In electronics, matrix-vector multiplication is done by charging wires, effecting them with transistors, and discharging them — spending energy at each step. In optical compute, input information is modulated directly on N mini-modulators and propagated through a designed weight mask, so weights modulate the light beam passively without spending extra active energy; the result is summed passively on detectors through physics and read out directly. Energy consumption correlates with the number of input/output units (N), notN

Y Combinator · What If We Stopped Using GPUs? | YC Paper Club

Claim from Ilker (optical computing expert)

Report · Money

The stimulation is very different from real biological signals, and the speaker bets it will keep diverging

The stimulation used is very, very different from real-life electrical signals you might get from your eyes and ears. Some people have tried training on real-life electrical signals from the brain, for example by growing mini eye organoids and attaching them to the brain — which the speaker finds scary because you'd be kind of recreating a human from first principles. The stimulation could evolve to a point where it starts looking similar to real biological inputs, and that might be a bit scary, but the speaker's bet is that it will continue diverging. What we will see is a different form of intelligence arise on the same substrate that we see intelligence on ourselves. Human intelligence is not optimal for compute; we're optimized for survival, reproduction, and the existence of our race.

Y Combinator · What If We Stopped Using GPUs? | YC Paper Club

Claim from Y Combinator

The speaker's bet about divergence is explicitly a prediction, not a fact.

Principle · Money

Attention is n-squared, driving exponential memory demand; GFLOP-per-joule has petered out

Memory capacity and bandwidth are on an exponential trajectory because attention is n-squared. Post-2020, compute efficiency is treated as acceptable as long as it produces the flops, even if it burns all the oceans. In the last two years there has been little improvement in GFLOP-per-joule, so something else is needed.

Y Combinator · What If We Stopped Using GPUs? | YC Paper Club

Claim from Y Combinator host