Links (3)
Links for the Links Throne! Articles for the Article God!
Disclaimer: Just because I include a link does not mean I endorse it or am confident in its claims. Unlike Scott Alexander, I don’t know the average number of links that will be proven wrong in each post because I have a much smaller readership that’s evidently too lazy to send me much-needed corrections derived from painstaking fact checking. (Hint, hint, but also in case it’s not clear, I’m just kidding. I genuinely cherish your scarce attention and occasional emails dearly.)
Jack Clark on Conversations with Tyler: Jack Clark! Tyler Cowen! As Tyler would say, “self-recommending”. Jack is one of the people whose opinions I trust most in the space, and this interview touched on topics he doesn’t necessarily cover in Import AI.
The group chats that changed America: As someone who likes group chats and finds them reminiscent of the early internet forums I have nostalgia for, I approve of the idea that doing participating in them is the key to the levers of power. That said, it does make me a bit sad how the open internet has retreated to private spaces.
Vibecoding MenuGen: The GOAT (Karpathy) on his experience vibe coding. The amount of schlep he experienced to actually deploy anything working matches my (limited) experience.
TLDR. Vibe coding menugen was exhilarating and fun escapade as a local demo, but a bit of a painful slog as a deployed, real app. Building a modern app is a bit like assembling IKEA future. There are all these services, docs, API keys, configurations, dev/prod deployments, team and security features, rate limits, pricing tiers... Meanwhile the LLMs have slightly outdated knowledge of everything, they make subtle but critical design mistakes when you watch them closely, and sometimes they hallucinate or gaslight you about solutions. But the most interesting part to me was that I didn't even spend all that much work in the code editor itself. I spent most of it in the browser, moving between tabs and settings and configuring and gluing a monster. All of this work and state is not even accessible or manipulatable by an LLM - how are we supposed to be automating society by 2027 like this?
The Intelligence Curse: I have to admit, for a long time I wasn’t as concerned about centralization as a major AI risk. It seemed too “squishy” somehow? If we have AIs that are aligned, won’t they be able to figure out this empowerment stuff?
Even before reading this, I had started to worry more, mainly because 1) there’s a set of legal / social innovations that would help here that will be very hard to get going preemptively (h/t Seb, who has been talking about this for a while) and 2) aligning AI systems to individuals seems technically more difficult than aligning them to aggregate preferences. Regarding 1, can a legal agent have attorney/client privilege with its principal? Not that I know of today. Should it be able to? Potentially! Is this something that’s even on regulator’s radar? Doubtful!
Anyway, I don’t agree with the entire series but I enjoyed reading it and am glad it’s out there.The Hard Thing #5: The Myth of Platform Biotech: “Some of you reading this think you are the one true platform that will change this dynamic and I am cheering you on, I swear!” I am in this picture and I’m ok with it!
Stepping back, I really enjoyed this post’s novel take on how things like the “magic hands” phenomenon are the bottleneck to solve for more scalable, shareable biology. Feels like a missing puzzle piece that fits nicely with my fast feedback loop maximalism.
Also, so many good people writing in the bio space these days! Dare I say we are entering a golden age of bio blogging? Remains to be seen.LMGTFY: A primer on therapeutics R&D: Also by Anna Marie Wagner, this one’s a Deep Research report on the therapeutics R&D process. I think this is the first time I’ve found myself reading someone else’s Deep Research report and benefiting from it. (Nice work!)
The Intelligence Consolidation: Over-simplifying, claims VC hedging across many foundation model players implies a bet against scaling. Not sure this holds for the VCs with big enough funds to be able to afford spanning most of the space, but still an interesting post.
At a portfolio level, the best case for the current spray-and-pray deployment of venture capital is case 2: indefinite optimism in some sort of AI progress, but without any clear conviction and with a long-term bet against scaling.
We would not bet against scaling! If you take scaling seriously, you end up in case 3: many of the enormous investments into frontier AI labs should be thought of as having the risk profile of early-stage investments but with the capital requirements of growth-stage investments. Diversifying on that basis isn’t prudent; it’s simply indecisive.
Leading from the Front: I’ve sung StaySaaSy’s praises before, so I won’t shill too much here. Another good, although not revelatory, post from them on what leading from the front looks like and means.
To learn to coordinate resources you must simply do it. Shirking away from hard product decisions, or arbitrating disagreements, or late night incidents will only entrench your lack of ability. Confidence in how to wrangle people, how to organize thought and action, and how to drive next steps is only learned from experience.
Second, your demeanor must be poised and determined and urgent. If you enter an incident hysterical or say “how the hell did this happen”, you are a distraction. Go fly a kite. Instead, your presence must imbue the team with a sense of “it does not matter how we got here. We will fix this and I will be here until we do.”
Impact, Agency, and Taste: Ben Kuhn goes from the first derivative to the second with a discussion of how to work on the right things and make the right things happen.
I think of finding high-leverage work as having two interrelated components:
a. Agency: i.e. some combination of the initiative/proactiveness to try to make things happen, and relentlessness and resourcefulness to make sure you’ll succeed.Taste: you need a good intuition for what things will and won’t work well to try.
b. Taste is important both “in the large” (picking important problems) and “in the small” (picking approaches to solving those problems that will work well); I usually see people first become great at the latter, then the former.
One of the easiest ways to get more leverage is to take a goal you’re already trying to accomplish, and figure out a better way to accomplish the same thing. (For example: changing your experiment design to be more directly relevant to the high-level question; finding an 80/20 way of building a tool; just deciding not to do something entirely; etc.) In order to do this, you need to keep careful track of the high-level context and priorities that your specific project is aimed at, so that you can understand and judge the trade-offs.
Tariffs & Strategy: Another realpolitik take on how to handle tariffs, this time from Roger Martin. Focuses specifically on Canada
This is not an economic fight that little countries can win. If they try to fight, they will learn the lesson Amex did with Costco. The two companies entered a long-term partnership in 1999 by which Amex became the exclusive credit card for Costco. By pretty much everyone’s definition, it was a gigantic success for both companies. But when the original deal came up for renewal in 2015, Costco competitively bid out the agreement. Amex was outraged, feeling that it had a partnership, not simply a supplier arrangement. Disgusted, Amex dropped out before the final round, which was won by rival Citibank/Visa.
Canada can hold its breath, stomp its feet, and say nasty things about the US and/or the Trump administration. But outrage is going to work as well for Canada as it did for Amex. Canada is small. The US is big. Canada needs the US market vastly more than the US needs Canada’s. Canada will have to come to some accommodation that works for both sides.
The future for all small-to-medium-sized countries — Canada included — is to increase the share of their trade with other similarly-sized countries, who have the same interests as they do. As always, WTP is an important strategy choice. For the better part of 80 years, Canada’s WTP for exports worked just fine. But it doesn’t anymore.
Unclogging your Decision Factory: Roger Martin reframes white collar work as about decisions.
First, think of your job as manufacturing decisions. The more efficient you are at producing high-quality decisions, the more valuable you will be — even if that isn’t in your explicit job description or KPI. There may not be a straight line to your success, but your boss will think of you as a person who gets the really important stuff done — even if your boss doesn’t understand how you make it happen.
Second, think of your job primarily as projects and secondarily as ongoing activities. Shrink the time you spend on ongoing, flat activities to the minimum possible so that you can dedicate your time to projects that require you to produce high-quality decisions. Don’t think of projects as onerous additions to your calendar but rather think of ongoing work as the task you need to cost/time-reduce to leave room for the truly important tasks that are central to the modern decision factory. And help your subordinates migrate in this direction too — you will be doing them a big favor.
Ask Art: What was Danaher like in the early days of Lean?: Stories from Danaher, one of my favorite titans about their early Lean Manufacturing days.
We started working with them in Jacobs Chuck in Clemson SC, which by the way also looked like a cave. We started with a plant tour which we thought would take about an hour and a half. We got only about 100 yards into the plant when they said, “Stop, we’ve seen enough.”
We returned immediately to the conference room, where Mr. Iwata went to the white board and in big letters wrote NO GOOD! He turned around and said: “Look, everything here is no good, what do you want to do about it?”
“I hate conveyer belts, get rid of them.” Yikes! The other team, with Nakao and Takanaka, started moving equipment right away (machines that had been in the plant for more than 15 years without being moved) and by the end of the day had created the first cell. The great lesson we learned here was to listen and learn. Don’t push back. We found right away that lean (Just-i-nTime back then) was “learn-by-doing” and we wanted to learn.
Intel’s Fall From Grace: Richard Rumelt, the author of one of my favorite nonfiction books, Good Strategy, Bad Strategy, is blogging again! This essay is about how Intel’s failure can be traced back to a coherent but outdated strategy and the challenge of adapting.
The dominant narrative in the financial press is that “financialization” damaged Intel. That is, a focus on stock buybacks and dividends rather than R&D hurt the company. This is basically wrongheaded as the story told here will demonstrate.
A second common narrative is that the board was filled with generalists who lacked understanding of the semiconductor business. This has been true, but was not the central issue.
A third common explanation is that bureaucracy strangled innovation. Intel certainly had bureaucracy, but was it the critical problem?
A fourth explanation is that Intel was “disrupted from below” by the ARM ecosystem which arose within the mobile space. This has an element of truth but does not explain Intel’s stumbles in graphics, AI, and its dramatic loss of leadership in its core CPU business in the 2016-24 era.
The hard truth is that Intel engaged in one of the world’s most coherent specialized and successful business strategies over almost three decades. It was this internally coherent specialization that led to its misses and failures.
Our Biotech Future: Freeman Dyson the future of biotech. This was a reread (thanks to Jeff for reminding me of it) but hit especially hard given some of the short term pain the biotech space is experiencing right now. Much has not yet come to pass, but perhaps it still may. In light of Colossal, Dyson seems prescient.
Humanoid Robots in Manufacturing: Ben Reinhardt does some analysis and comes out skeptical of humanoids for use in manufacturing (but bullish for other uses).
Humanoid robots will be like microprocessors: a general purpose tool that is so much cheaper and easier to use than its specialized counterpart that most people use it, despite not being optimal. In the world of microprocessors, you can create a custom chip that can do almost any specific computing task better than a general-purpose chip. However, there are only a few applications where the time and cost to create a custom chip is worth it. One could imagine a world where the same is true for robotics. While I can’t rule this out, I think there are some key differences between robots and microprocessors that make this scenario unlikely. The vast majority of the cost of a chip is in the design and initial tape-out, so there are huge economies of scale: you need to make many of that specific chip to make the cost per chip reasonable. On the other hand, a lot of the cost of a robot is in the components – the motors, sensors, and raw material. These components are general purpose – as you drive down the cost of the motors for a humanoid robot, you also drive down the cost of other robots that also use motors.
Humanoid robots being general-purpose and easy-to-train at all assumes that we have much better robotic software. (I realize I sound like a broken record at this point.) It’s likely that the same software that makes humanoid robots work would also make non-humanoid robots work. It would also be a strange world where we have software that enables humanoid robots to easily do most work but we don’t have software that can help design better specialized systems, or teams of specialized robots.
In high-volume situations, it already makes sense to build specialized tools that can produce widgets far faster than humanoids. The same technology that we need to make useful humanoids will drive down the volume that makes sense for specialized machines. As the gap between human and robotic capabilities closes in low volume situations, the same software and hardware that will enable useful humanoid robots will also enable cheaper, more robust, and faster automated systems that are just harder to imagine right now.
How to Fix Crime in New York City: Or really, how crime was previously reduced in New York City. I seem doomed to include one or two vaguely political links in each links post… What I liked about this one was not any political connotation or inferred relevance to now but its discussion of what, in practice, what stopping crime involved. As has been my own experience with everything on which I’ve worked, it’s often the little, messy details that count. Lots of that here.
Will you describe how they tackle the squeegee men problem?
Mike Julian was behind it. They hired George Kelling, who's known for broken windows. They said, “These people are here to make money. So to just go there and make a few arrests isn't going to solve the problem.” First of all, he had to figure out what legal authority [to use], and he used Traffic Reg 44 [which prohibits pedestrians from soliciting vehicle occupants]. He talked to Norm Siegel of the NYCLU [New York Civil Liberties Union] about this, who did not want this crackdown to happen. But Norman said, “Okay, this is the law, I can't fight that one. You're doing it legally. It's all in the books.” And So that took away that opposition.
But the relentless part of it is key. First they filmed people. Then, when it came to enforcement, they warned people. Then they cited people, and anybody that was left they arrested. They did not have to arrest many people, because the key is they did this every four hours. It was that that changed behavior, because even a simple arrest isn't going to necessarily deter someone if it's a productive way to make money. But being out there every four hours for a couple of weeks or months was enough to get people to do something else. What that something else is, we still don't know, but we solved the squeegee problem.
But as you flesh out, the meat on the bones of CompStat was this relentless follow-up. You'd have these weekly meetings early in the morning with all the precinct heads. There were relentless asks from the bosses, “What's going on in your district or in your precinct? Can you explain why this is happening? What are you doing to get these numbers down?” And follow-ups the following week or month. It was constant.
On Prions and Protein Design: Eryney and Theo write in Asimov about how proteins as a model system for computational dynamics prediction. Figuring this out is also important because it’s the only way we can engineer the dracomen from the Polity.
Mental Strength in Judo, Mental Strength in Life: First of five essays by Cedric Chin that I read this month. In this one, Cedric writes about his experience spending several months training nearly full time for a judo competition. For me, the most interesting part was on the tendency to intellectualize will issues as skill issues. Obviously not relevant to me personally, someone who never overthinks things and definitely did not have a problem when he played soccer of watching the ball and thinking about it rather than kicking it…
“Yes, yes, never mind that,” he said, waving me away. “But have you ever considered, what’s the assumption that someone makes if they say they have no talent?”
I kept quiet.
“It means that they think their problem is technical. They believe they do not have the skills.”
“Oh,” I said.
“And so I thought that your problem was technical! I trained you as if your problem was technical. But it wasn’t! No matter what I taught you, you didn’t seem to improve. Your problem is a matter of will, not a matter of skill. I taught you everything you needed to know in the first two months, but you just didn’t seem to apply it! You were not willing to apply it.”
I kept quiet, so he continued. “And so I think here’s what happened. You think that your problem is technical, so you kept telling yourself ‘oh, I can’t do the seoi overshoot, or I can’t outgrip A, or I lost to J because they know more than me, or they have more skills than me.’ But that is not true. You are giving yourself excuses! Do you think that when A is thrown by you or pinned by you, he thinks ‘oh, Cedric beat me because he’s more skilful or he’s stronger?’ I can bet you my last dollar that he doesn’t! I guarantee you he’s thinking something like ‘goddammit, he got me that one time, but I’m never letting him get me that way again!’ But you do not. You do not think like this, do you?”
Business with Tariffs; Business as Usual: More Cedric! This time on tariffs and how working within their constraints is a reversion to historical norms. I definitely benefit from a good ol’, “stop acting like the sky is falling and work within the constraints” pep talk, and this acted as that.
But imagine that. Imagine growing up under one empire, one world order, one trade currency, and watching all of that unwind as your career begins properly. Today, a non-mainstream, crank-adjacent opinion is “we are watching the death of the US dollar”, and folks in most of my social circles sort of ignore it. I’m not saying that a decline of the US dollar will or won’t happen, only that it’s not unimaginable — here we have one man who saw the global trade currency and the global reserve currency change before his eyes before he hit middle age. If you were to ask him “what would you do given the uncertainty around tariffs today?”, Kuok would likely shrug as he trimmed his bets. He’s seen worse before. He wouldn’t be surprised.
Vanguard as a Demand Side Mystery: Why yes, I do like Cedric’s writing. How can you tell?
Inspiring story of the birth of Vanguard and the index fund industry. Very much a marathon, not a sprint that ended up being a bet on compounding and the market being a weighing machine in the long run. Interesting because, "this is not typically how a large company becomes large" and yet Vanguard is really freaking large.The Heart of Innovation: Why Most Startups Fail: Follow up to the prior post. Summarizes The Heart of Innovation, which proposes a new theory of demand that explains what theories like, “people buy things to remove pain” do not. I tried to read the book and found it tough going, but Cedric’s summary helped me wrap my head around the key ideas and now I feel like I could go back to the book and get more out of it.
The Tricky Thing About Creating Training Programs: Last post from Cedric this month…
And if there’s one big insight there, these performances and the training of them tend to be very focused on what you do, and yet the NDM-type research that you come across, the higher you go in the expertise tree, the more it becomes about what you see, instead of what you do. So we’re going to push that aside, the ‘what you do’ bit — we’re not saying it’s not important, we’re saying it’s already addressed (in current training methods). What’s under-addressed and what gives us the opportunity to perhaps accelerate the development of expertise is this focus on what you see. And it turns out that applying some of these video-occlusion type things can be done on a drill-and-practice level, and they can be done on an iPhone. So we can get a lot of efficiencies out of that kind of approach.