“Sternin, you have six months to show results.”
A senior official in Vietnam’s Ministry of Foreign Affairs said it to Jerry Sternin, who had arrived in Hanoi in December 1990 to open an office for Save the Children. His job was child malnutrition. By the account he later helped write, two-thirds of the country’s children under five were malnourished. If Sternin couldn’t show impact in six months, the official couldn’t extend his visa.
The government had already watched handouts of extra food fail to last. When the programs ended, the gains tapered off. Sternin and his wife, Monique, had little money and no way to make the villages richer by summer.
In four villages of Quang Xuong district, health volunteers weighed about 2,000 children under three. Some 64 percent were malnourished. Then the Sternins asked a question that had nothing to do with money. Were there any well-nourished children from very, very poor families?
Yes, came the answer. There were some.
What the poor families already knew
Those families were doing things anyone in the village could do. They added sweet potato greens to their children’s food and caught tiny shrimps and crabs in the rice paddies. They fed the children three or four times a day instead of the usual two, fed them by hand so nothing was wasted, and washed their hands before and after meals. The food was free for the taking. Most families believed it was wrong for small children.
Mothers of malnourished children waded into the paddies with nets and cooked the new dishes together for two weeks, weighing the children as they went. Six months after the Sternins arrived, the official handed them a renewed visa.
Spend your effort on what you can reach, not on everything you worry about.
Stephen Covey drew this as two circles in The 7 Habits of Highly Effective People, the 1989 book that went on to sell more than 20 million copies. The big circle, concern, holds everything you care about: the economy, the weather, other people’s choices. Inside it sits a smaller circle, influence, the part you can actually do something about. “Proactive people focus their efforts in the Circle of Influence,” Covey wrote, as the FranklinPlanner blog quotes him. “They work on the things they can do something about.” His claim is that effort spent there makes the inner circle grow over time, while people who dwell on the outer ring feel like victims and see their influence shrink.
The Sternins’ circle did grow. By their account, malnutrition in the pilot communities fell 85 percent over two years, and the approach became a national program reaching more than 2.2 million people.
Where the circles stop
The circles are a framework for sorting your attention, not a research finding, and they leave out two things.
First, what sits outside your circle still matters. The Vietnamese villages were still poor. The method also depended on something lucky: some poor families had already found an answer. Where nobody has, there is nothing nearby to copy.
Second, a story that ends well is not proof. Writing in the BMJ in 2004, the Sternins and their colleagues reported a 74 percent drop in severe malnutrition among participating children under three in the early evaluations, and said plainly that those designs were too weak to show the program caused it. So Save the Children ran a randomized trial in 12 communes in northern Vietnam. The children in the program communes grew better.
The budget that got cut
Imagine you lead a team of eight, and head office has just cut next year’s budget by a fifth. You think the decision is wrong, and you may be right.
One message to your director, with the numbers and a proposal, is inside your circle of influence. You can’t decide the budget, but you can change what the person who decides it knows. Two weeks of the same complaint, repeated in the team-lead chat with nothing new in it, is not. It changes nothing anyone knows.
Inside the circle there are also smaller, duller questions. Which two projects stop first? Which other team ran on a similar budget last year and still shipped? What did they drop?
Those questions won’t restore the money. They will decide what your team delivers with what’s left. Send the one case worth making, then spend the rest of the fortnight on them; the quarter has already started.
The model that changed under you
If you build on an AI model through an API, the model itself sits in someone else’s circle. In July 2023, researchers at Stanford and UC Berkeley compared the March and June snapshots of GPT-4. In the first version of their paper, GPT-4’s accuracy on a set of 500 numbers fell from 97.6 percent to 2.4 percent when asked whether each was prime.
Arvind Narayanan and Sayash Kapoor, writing on their blog AI Snake Oil, pointed out that all 500 numbers were prime. When they tested composite numbers too, the models mostly guessed: the March GPT-4 nearly always said “prime,” the June one nearly always said “composite.” Behavior had changed, not necessarily ability. For anyone whose product depended on the old behavior, they noted, the difference was little comfort.
This is where the circles need adapting. With people, some things are simply out of reach. With a model, part of the outer ring can be pulled inside by setup. You can call a dated snapshot instead of a moving name, though Narayanan and Kapoor noted in 2023 that OpenAI kept snapshots for only a few months. You can keep a test set of your own real tasks, with both kinds of right answer in it, and rerun it whenever the model changes.
The test set won’t stop the next change. It can show it to you on your own cases first, as long as your cases include the composites.