Notes

As John Ousterhout puts it, a little bit of slope makes up for a lot of y-intercept.

Two careers: one starts ahead, the other learns faster and passes it

Time	intercept	slope
0	4.25	1.46
6.5		10
10	6.33

Ramp hires for slope by name, and Jane Street calls a finance background optional. Traditional interviews miss slope, part of why I prefer work trials. It's contagious, too: teachers improve faster when their colleagues are better. That is the real case for talent density. A company's curve is the sum of its people's, so collective slope sets how fast it grows.

Zoom in, and each of those people is a line of skills, with similar skills sitting close together. Peaks are strengths and valleys are weaknesses. Many peaks, far apart, make a generalist; one or two make a specialist.

Ability

specialistgeneralistexceptional generalistfar apart: it barely fillsnearby peaks: the valley fills

Skills (similar skills sit close together)

A specialist, a generalist and an exceptional generalist, with two valleys filling in

Skills (similar skills sit close together)	specialist	generalist	exceptional generalist
0	-0.17	-0.03	0.19
0.1	-0.27	0.45	0.19
0.2	0.18	0.01	-0.14
0.3	-0.1	-0.23	-0.3
0.4	-0.26	0.01	-0.15
0.5	-0.46	-0.02	0.76
0.6	-0.07	-0.19	0.66
0.7	-0.1	0.03	-0.22
0.8	-0.47	0.23	-0.17
0.9	-0.21	-0.03	0.56
1	-0.13	0.16	-0.09

The valleys between nearby peaks are the easiest to fill. In my experience, high-slope people fill their valleys faster. Working with people who peak where you want to grow fills valleys and extends peaks.

The best leaders know each person's peaks, and the valleys they want to grow. They scope work to peaks, give bonus scope in valleys, and expose them to the right people. High-slope people take on more of that bonus, because they simply move faster.

Quality: how well a product performs when the owner uses it. Experience: how well a set of products perform for the owner at the same time.

Most consumers maximize quality/price or just minimize price. This makes sense since money is finite and there is comfort in frugality. However, why follow this approach for things that you deeply care about?

Time is important to me. So I've started optimizing experiences that give me more time. In doing so, my purchasing has become experimental. I've followed roughly these five steps to optimize experiences:

1. Identify its constraints.
2. Define the experience.
3. Measure a baseline.
4. Buy the candidate products and test them in combinations.
5. Keep the product set that maximizes the experience.
6. Return or sell everything else.

Consider my experience sleeping. Constraints: I can only fall asleep listening to podcasts. Sound and light wake me up. Sound and light wake up my partner. I sleep on my stomach with my ear on the pillow. Define the experience: My sleep, measured by sleep efficiency. The time in bed I spend awake is my loss function. It depends on the set of products I sleep with, and the goal is to find the set that minimizes it.

Measure a baseline: I measured with my Whoop, a voice memo of what woke me, and a morning note to confirm. On average, I took 26 minutes to fall asleep and lay awake another 35 between then and my alarm.

Noise cost 29, light 16, ear discomfort 6, the podcast 3, and 7 were no product’s fault. The split by cause is my best guess, so treat it as rough and the total as solid. Run experiments: Each set I tried was a hypothesis, tested against the baseline of 26 + 35 = 61 minutes lost.

1. AirPods + eye mask: 24 + 20 = 44 minutes lost. Podcast and noise cancelling in one device, but ear discomfort kept me awake.
2. Loop earplugs + podcast on my phone speaker + eye mask: 10 + 6 = 16 minutes lost. The earplugs are flat enough to sleep on and block the sound that wakes me. With the phone close to my head and the volume up, the podcast still gets through them.
3. Manta sleep mask alone: 17 + 16 = 33 minutes lost, so its quality is $1 - 33/480 \approx 93\%$. One device for sound, light and the podcast, but very uncomfortable for stomach sleeping.
4. Sleep earbuds + eye mask: 15 + 20 = 35 minutes lost. Built for sleeping on your ear, very uncomfortable and poor noise cancellation.

Keep the product set that maximizes the experience. B won, but it broke a constraint: my partner could hear the podcast. The fix was a second pair of Loop earplugs to prevent her from hearing my podcast. With the phone by my head, the podcast gets through my earplugs but not hers. Return or resell everything else. I returned or resold everything else I bought except my Loop earplugs and my best eye mask. Results: The loss function fell from 61 minutes to 16, so sleep efficiency is $1 - 16/480 \approx 97\%$ (was 87%). Over a year, that’s $45 \text{ min} \times 365 \approx 274 \text{ h}$ more sleep, about 11 days. Counterintuitively, the best experience had the cheapest products. All this said, I’m not going to run trials on socks. Most purchases don’t deserve an experiment, and I don’t have the time to give them one. But for the experiences that impact my time, I’m starting to experimentally purchase. Otherwise I’m surrounding myself with expensive objects and hoping.

Radioactive materials have half-lives. What if beliefs do too?

Most of us are trained to keep our half-lives long. School awards correctness. When I take up a belief, I think it's true. Changing this belief is admission I was incorrect. Social pressures extend half-lives. In a Pew survey, +80% of teens inherit their parents' political party. Sunk costs stretch half-lives further: years acting on a belief feel wasted if you drop it. At the same time, new experiences naturally challenge the stability of convictions.

So how do you ensure adequate half-life calibration? I don't know. Right now I ask myself: Am I prioritizing correctness over truth? yes -> minimize half-life Am I confident my experience will later prove my conviction? yes -> maximize half-life

Organizations face a version of this too. Patrick Collison writes about replacement rate. Interestingly, the mind is different from an organization. When a company fails, the market learns; when a person's convictions fail, only they can.

Alaska's halibut season shrank from 122 days in 1979 to two or three days by 1991. Any fish one boat left, another would catch, so every boat raced. They called it the derby: crews fished in gales, and maydays swamped the Coast Guard.

Apps race for your attention the same way: any hour one app leaves, another takes. Netflix's CEO once said they were competing with sleep. Meta makes about a dollar per hour of American attention, and pays nothing for the catch. American teens give social media 4.8 hours a day. In 2012, 27% of 13-year-olds read for fun almost daily; by 2023, 14% did.

Students in one study wanted $55 to quit TikTok for a month on their own. If their whole campus quit too, they'd pay $24 to make it happen. It's a derby: nobody can slow down alone. Alaska ended its derby in 1995 by giving boat owners shares of a capped catch. With nothing left to race for, the season stretched to eight months.

I'd end the attention derby with a tax on every hour a platform catches. The economists behind that TikTok study say their results could justify a sizable one. The rate would climb with each hour of your day, faster for kids. Once your fifth hour costs more in tax than it earns in ads, apps stop racing for it. And the money would go to schools, to restock the readers the race fished out.