Tim retweeted
Jamie Dimon put the commodity question straight to Dario Amodei.
Dimon: "You don't need expensive models. You can use far cheaper models, cheaper tokens to do the calculations"
"Where would that be wrong?"
@DarioAmodei: "every time we extend the frontier, it creates entirely new markets." He went on: in 2023 everyone was focused on chatbots, and Anthropic bet that coding would be much bigger.
He's right that each jump opens a new market. The catch is in Dimon's question: every new market gets its cheaper model eventually, so the frontier pays only while new markets arrive faster than old ones commoditize.
From @jpmorgan's investor livestream, days before @AnthropicAI's IPO.
Tim retweeted
Whatever you make of this week's superintelligence timelines, the useful question for a company is a smaller one.
Which parts of your work would you hand to a model twice as good as today's, and what has to be true first? Clean data, a named owner, a clear test for what done looks like.
Write that list now. If the forecasts are right, it's your roadmap. If they're wrong, it's still your roadmap, on a slower clock.
Michael Rubin says he came within hours of losing his company to a margin call in 2009. In March 2011, @eBay agreed to buy it for $2.4 billion.
The company was GSI Commerce, a public company that ran online stores for big retailers and brands, and @michaelrubin held a large stake in it. By his account to @davidsenra, he had been borrowing against that stock for years and putting the cash into the market. He says the shares were worth about $250 million in 2007, then fell about 90% in the financial crisis, and the bank's lending limits fell with them.
So the loan had to come down:
"I owed $50 million to one of the big banks, which I got down to $3 million personally."
If he couldn't find the last $3 million, the bank would start selling his shares. He found it "with hours to spare, not days. Hours."
Two years later eBay bought GSI, and as part of that deal its sports merchandise business, @Fanatics, went to a new company Rubin led. He now puts Fanatics at $14 billion of revenue this year and 22,000 employees.
His explanation for getting through 2009: "I was never frazzled. I was never nervous. I just kept working."
The usual lesson here is grit. Mine is about the loan: he was borrowing against shares of his own company, so one crash hit the business and the collateral at the same time. A founder's margin loan is a second bet on the same company, made with a bank's patience.
From @davidsenra's conversation with @michaelrubin.
2
1
517
We book meetings for B2B companies on a pay-per-meeting basis.
You only pay for qualified prospects who actually show up.
If you're interested, book a call with us here:
introslot.com/revenueflow/ca…
113
Tim retweeted
Jamie Dimon says JPMorgan first saw AI in 2012, when it hired @PalantirTech.
"And we now have thousands of people doing it, aided and abetted by Anthropic"
His test for all of it: "we are going to use AI to do a better job for our clients. That's it."
I'd take that one-line test over a 40-page AI strategy deck, because usage numbers are easy to report and say nothing about the client.
From @jpmorgan's investor livestream with @AnthropicAI CEO @DarioAmodei.
Jamie Dimon put the commodity question straight to Dario Amodei.
Dimon: "You don't need expensive models. You can use far cheaper models, cheaper tokens to do the calculations"
"Where would that be wrong?"
@DarioAmodei: "every time we extend the frontier, it creates entirely new markets." He went on: in 2023 everyone was focused on chatbots, and Anthropic bet that coding would be much bigger.
He's right that each jump opens a new market. The catch is in Dimon's question: every new market gets its cheaper model eventually, so the frontier pays only while new markets arrive faster than old ones commoditize.
From @jpmorgan's investor livestream, days before @AnthropicAI's IPO.
We book meetings for B2B companies on a pay-per-meeting basis.
You only pay for qualified prospects who actually show up.
If you're interested, book a call with us here:
introslot.com/revenueflow/ca…
125
Whatever you make of this week's superintelligence timelines, the useful question for a company is a smaller one.
Which parts of your work would you hand to a model twice as good as today's, and what has to be true first? Clean data, a named owner, a clear test for what done looks like.
Write that list now. If the forecasts are right, it's your roadmap. If they're wrong, it's still your roadmap, on a slower clock.
We book meetings for B2B companies on a pay-per-meeting basis.
You only pay for qualified prospects who actually show up.
If you're interested, book a call with us here:
introslot.com/revenueflow/ca…
129
Jamie Dimon says JPMorgan first saw AI in 2012, when it hired @PalantirTech.
"And we now have thousands of people doing it, aided and abetted by Anthropic"
His test for all of it: "we are going to use AI to do a better job for our clients. That's it."
I'd take that one-line test over a 40-page AI strategy deck, because usage numbers are easy to report and say nothing about the client.
From @jpmorgan's investor livestream with @AnthropicAI CEO @DarioAmodei.
We book meetings for B2B companies on a pay-per-meeting basis.
You only pay for qualified prospects who actually show up.
If you're interested, book a call with us here:
introslot.com/revenueflow/ca…
179
Dario Amodei says AI now finds security holes faster than companies can patch them.
Jamie Dimon asked @DarioAmodei how companies should get ready:
"Our customers have found hundreds of thousands of vulnerabilities. We found hundreds of thousands of vulnerabilities. Then the limiting step becomes, how do you actually patch the vulnerabilities?"
@AnthropicAI opened its Mythos model to partners for defensive work in April. He says @claudeai is getting better at the patching half.
Finding a bug got cheap. Fixing one still takes someone who knows the system, and attackers can run the same models on the same code. The patch backlog is where the next breaches come from.
From @jpmorgan's investor livestream with @DarioAmodei.
We book meetings for B2B companies on a pay-per-meeting basis.
You only pay for qualified prospects who actually show up.
If you're interested, book a call with us here:
introslot.com/revenueflow/ca…
202
SpaceX caught Starship with the tower for the first time. Just over an hour later, @elonmusk posted: "Inspecting ship 42 right now."
Flight 15 deployed 26 @Starlink V3 satellites, and Ship 42 came back to Starbase about 22.5 hours after liftoff. The booster was diverted to an offshore splashdown.
The ship is the hard half of reuse because it comes back from orbital speed. If SpaceX can do this on repeat, the price of reaching orbit is the next number to fall.
Video: @SpaceX.
We book meetings for B2B companies on a pay-per-meeting basis.
You only pay for qualified prospects who actually show up.
If you're interested, book a call with us here:
introslot.com/revenueflow/ca…
162
Status symbols at AI startups right now:
- A token bill you can explain line by line
- Evals you wrote before the demo
- Agents that ask before they spend money
- More agents in production than slides about agents
- Customers who never mention it's AI
- A roadmap that lists what you decided not to build
We book meetings for B2B companies on a pay-per-meeting basis.
You only pay for qualified prospects who actually show up.
If you're interested, book a call with us here:
introslot.com/revenueflow/ca…
151
Dario Amodei says Anthropic buys "relatively little compute" ahead of time. Its IPO filing lists nearly $900 billion of compute and data center commitments.
Both are true, and the gap between them was the most interesting answer he gave on Friday.
@DarioAmodei sat down with Jamie Dimon at a livestreamed @jpmorgan investor event, days before @AnthropicAI is expected to start trading on Nasdaq. Dimon brought up the roughly $900 billion of forward commitments Anthropic has disclosed and asked how it knows it will get the compute it needs and keep it cost-efficient.
Amodei called compute "the deepest business dilemma that the company faces." Revenue grew roughly 10x in each of 2024 and 2025, he said, and every year nobody inside knows what the next one brings. But compute has to be bought a year or two ahead. Buy too much and you can overextend yourself. Buy too little and you can't serve customers or train the next model.
His answer:
"The way we think about that is we actually tend to buy on the low side. We buy relatively little compute for relatively low cost ahead of time."
If demand beats the plan, Anthropic tops up on the spot market, "which so far has always been available and which we believe will continue to be available. That costs more, but we're buying it in worlds where we're doing particularly well."
"So you can think of it a little bit like we're buying an insurance policy."
What "relatively little" looks like on paper:
→ About $895 billion of purchase commitments and leases in its amended S-1, nearly all of it compute and infrastructure
→ Roughly half of that due before the end of 2030
→ The IPO itself: $70 to $78 a share, up to about $64.4 billion raised, with trading expected around November 19
So the low side means low compared with what Anthropic could need in its best year.
My take: the hedge is smart, and the weak spot is the sentence about the spot market. A great year for Anthropic is very likely a great year for every AI lab, so the day it needs extra compute is the day everyone else is bidding for the same chips. Normal insurance pays out when things go wrong. This one has Anthropic paying top prices exactly when it is winning, which works as long as winning comes with the margins to cover it.
From @jpmorgan's investor livestream with @DarioAmodei.
We book meetings for B2B companies on a pay-per-meeting basis.
You only pay for qualified prospects who actually show up.
If you're interested, book a call with us here:
introslot.com/revenueflow/ca…
182
Dwarkesh Patel thinks Claude 7 could have invented the Transformer.
"Even the Transformer... is the kind of thing that Claude 7 could have invented if Chris Olah asked it to try really hard to get this attention thing to work better."
@ch402 works at @AnthropicAI, so this one can be tested in-house: give a model only the papers from before 2017 and see what it builds.
From @dwarkesh_sp.
We book meetings for B2B companies on a pay-per-meeting basis.
You only pay for qualified prospects who actually show up.
If you're interested, book a call with us here:
introslot.com/revenueflow/ca…
327
