AI in action

One prediction/analogy of frontier models

Why the frontier labs subsidize so heavily

How businesses should charge in the world of AI

Atoms has been operating in stealth for years, mainly focusing on 'cloud kitchens' for food preparation, as well as more recently automated mining via an acquisition. Atoms is the evolution of Uber in a sense, with a mission of 'digitizing the physical world'

“The idea is an ‘atoms-based computer’:

1. CPU manipulates bits. What manipulates atoms - that’s manufacturing
2. Storage stores bits. What stores atoms - that’s real estate
3. Network moves bits. What moves atoms - that’s transport logistics

Those are your three core computing resources in  an atoms based computer. You can model just about anything with computation.”

“The idea of treating atoms like bits started at uber. It was science fiction at one point: you would touch glass (your phone) and a car would come to you, and you had satellite view and you’re watching this car come to you, it was crazy. Now it’s just like whatever, but it was crazy at the time, everyone’s first Uber experience was a magical moment.”

"The meaningful thing about Atoms Food: can you make the preparation and delivery of a meal so efficient that it approaches the cost of going to a grocery store? If it does, you do to the kitchen what uber did to the car. That’s a big deal, but what do you need to do to get there? 3 things:

• Warehouses like Amazon style Picking, packing, sorting, and then putting in a car (we already have many hundreds of facilities in 30 countries)
• Robotics automated food/meal production (we’re getting a manufacturing line for our robots up this year, and essentially everyone in robotics would agree eventually robots will make better food than most people, and now you’re taking over $5 out in labor)
• Robotic couriers that hold food at temperature and come to your house (once we get the robotic couriers you go from over $10 a drop for each meal to under a dollar)

And you save a few bucks on occupancy per meal. And all of a sudden you are getting a high quality meal delivered to you all in for under $10 bucks. Based on robotic process and seeing Waymo’s already navigating autonomously, we will easily be here in 20 years. Can we do it faster? What about in 10, or in 7, or in 5? That’s when you’re bending reality, you’re bending it towards now. Everybody would agree that it’s going to happen, and now it’s about when and who. And then you say LFG."

Travis on the expanding into autonomous mining:

"More productive mines to power Earth’s industries. We’re operating in a bunch of mines around the world. When you automate certain parts of mining you not only get more gold or silver or whatever you’re mining, you dramatically improve safety in a real way. And new mines are going to emerge that were not viable before. 

Once you get the foothold and it’s working, you can keep expanding. Adjacent categories are way easier than whole new off the rails categories. First it’s uber rides then it’s uber eats then it’s uber autonomy then it’s uber freight then it’s uber AI labs. When it comes to expansion, you can start letting your imagination, with constraints, start to do really interesting things."

• Don’t die: Goldman survived 150 years, including depressions and crises and scandals and potential takeovers. Many once thriving firms did not (Lehman Brothers, Bear Sterns, Solomon Brothers, Long Term Capital Management). It came close a few times though…

• The top surprise: perhaps the 3 most crucial heads of the firm all took multiple years, even a decade, to gain real recondition:
• Sidney Weinberg: started as a janitor and mail clerk in his teens, eventually became a low level commercial paper salesmen for many years before eventually rising up and eventually leading the firm for 3 decades
• John Whitehead: “I was only serving time, not learning much, in my first years at Goldman.” and “I’d been at GS for 8 years and no one had even mentioned me being a partner”
• Lloyd Blankfein: barely avoided layoffs in his second year; when he eventually was given a major business to run years into his tenure within a few weeks they rescinded the leadership role (though he eventually earned it back)

• It was (at one point) generally on the side of the underdogs. It worked with entertainment and technology and airline companies because the top investment banks wanted to work mainly with railroads abs manufacturing companies. When hostile takeovers became much more permissible in the 1980s, it was the only major bank that refused to support companies trying to do the takeovers (only working with those trying to avoid or navigate being acquired) - this led to gain great trust with smaller firms that eventually became leasing ones.

• Recruiting was tremendously prioritized: Partners spent a lot of time recruiting, not just junior folks. In one year a partner interviewed 900 candidates from 35 schools for his department and selected just 23. There were very high standards to get in, and lots of training (both formal and informal) once you were in. 

• And this: Said an employee of Goldman Sachs in the early 90s: “Why so intensely engaged, so early in the morning every day? Because that’s what we truly wanted to do, and where we wanted to be. And that’s the way it was all the time for everybody at Goldman Sachs. At Morgan Stanley where I also worked for awhile, people saw their work as personally defining - it was what they could do and did do. But at Goldman Sachs it was much more. It was life.”

A few other excerpts from the book:

• Day after day Hank Paulson (CEO in early 2000's) took as his first priority completing whatever loomed as the one most difficult or unpleasant task on his to do list
• It was the decision of Paulson alone at the top not to combine Goldman Sachs with JP Morgan
• More than any Wall Street firm, the partners at GS turned up for weddings, funerals, and bar mitzvahs, and the people who left were still kept as close key members
• One business opens up opportunities for additional business lines. Happened time and time again for GS and was critical for its growth into the from it became. Commercial paper -> Investment Banking -> Block Trading / Market Making -> Arbitrage -> FX/Commodities -> Proprietary Trading -> Asset Management
• John Whitehead (one of the most crucial leaders of the firm who turned it to top tier investment bank): “we had no big and bad ideas…we knew it would take a generation to complete the change of our position. Doing thousands of little things day after day, inching along as consistently as you can in the right direction as best you can tell…as we made changes almost continuously we had many many failures."

Up until the last year, developing an app took enough time and expertise that it tended to be done for multiple people and repetitive use cases.

But AI coding capabilities have gotten so good that in literally a few minutes, you can make a tailor made app.

I tried out doing just that, using a train ride to make an apartment comparison app for one time use, just for me - few snapshots below. 

This month I started using ChatGPT to send me a daily morning briefing. I chose the specific set of sections and style I wanted, and have been able to iterate on the format. I'm finding what is essentially my own morning newsletter to be a great addition to my news routine, and recommend creating your own