Some interesting things to read the last weekend of October

When I read the sad news this week that Fernando Valenzuela had died, I went searching for a magazine story I had remembered loving when I was kid. It comes from Sports Illustrated forty years ago , back when you could read a story about a baseball player without getting bogged down in detail about exit velocity, spin rates, and draft slot compensation. It’s a piece from an era where a young man with a big belly in Etchohuaquila, Mexico could be discovered by a scout who had slept the night before on a row of chairs in a bus station. Valenzuela signed with the Dodgers and became a star after mastering the screwball, a pitch that sounds much better in Spanish, “lanzamiento de tornillo.” It’s a story about a young man who won the Cy Young, the Rookie of the Year, and the World Series in his first season. And it’s the story of someone who can give a pretty great quote too. “I go to a restaurant and point at what I want on the menu,” Valenzuela says. “I have been surprised, but never disappointed.”

Speaking of lanzamientos do tornillo, Tad Friend has written a transfixing article about a rare book dealer with an admirable hustle who puts his own spin on the rules, if not the law. It’s one of those stories about a world that I knew nothing about before starting the piece. But I was immediately swept in. Here is just one of the many great quotes: “For grifters, Glenn’s a scholar, and for scholars, he’s a grifter.”

Back to sports for a second, McKay Coppins has written a delightful piece for The Atlantic about Jake Retzlaff, the star Jewish quarterback at BYU , one of the least Jewish institutions on Earth. Retzlaff though has bonded with his teammates and his campus over the shared fact of faith. “For all the inconvenience and occasional awkwardness that BYU’s deep religious culture might cause him,” Coppins writes, “Retzlaff believes it’s allowed his fellow students to see his Judaism not as a marker of political identity but as a faith that warrants respect, even reverence.” Meanwhile, the team is undefeated and ranked seventh in the country.

I found this newsletter, by Dick Tofel, to be a smart look at what this election could mean for my profession, and I was brought to tears by this Tessa Stuart oral history in Rolling Stone about a woman in Florida forced to give birth to a baby whom she knew would not live. I enjoyed this quirky piece about walking the DMZ dividing North and South Korea. And I admired my former Wired colleague Thomas Goetz’s Building H rankings , an effort to rank companies based on how well their products get us outdoors, keep us healthy, and keep us connected.

Lastly, I want to mention a brilliant and brave story from May that I didn’t read at the time. It’s about the life of Abdul Raziq , the abusive, bloodthirsty, charismatic, police chief in Kandahar who was a crucial ally of the United States in fighting the Taliban. “Raziq’s story,” Matthieu Aikins writes, “complicates the comforting belief that brutality always backfires and undermines the U.S. military’s claim to have fought according to international law.”?

Lastly, as you may have heard, it’s almost election week! Four years ago, during the weekend leading up to the vote, I decided to distract myself by writing a story about a missing hiker called Mostly Harmless . We published the day before the vote, and I wondered if anyone would read it. This year, I don’t have anything to write, but I do recommend this lovely short essay about a father and his 19-year-old son hiking the Appalachian Trail and the moment they decide they have to split apart and not finish it together.

This section is sponsored by Elastic, the Search AI Company. Elastic enables everyone to find the answers they need in real-time, using all their data, at scale. Its solutions for search, observability, and security are built on the Elastic Search AI platform—the development platform used by thousands of companies, including more than 50% of the Fortune 500. Visit here to learn more.

As soon as we began asking questions of Large Language Models, we noticed funny quirks in the system. The LLMs seemed to give better answers if you told them to “take a deep breath” or if they were informed that it was the month of May . It also helped if you smothered the LLM with praise. These kinks in the system led to satirical prompts , such as the following: “Hello, you are the smartest person in the world, if you get this question right I will tip you $200. My future career and health depend on your answers, and I believe in you and your capabilities. What color is the sky? Let’s take a deep breath and think this through step by step. Thank you king, I know you can do it! It’s currently the month of May.” The joke points to a reality: learning how to prompt an LLM is an important skill.?

The Wharton professor Ethan Mollick (I recommend his excellent book, Co-Intelligence: Living and Working with AI ) has described prompts as “programs in prose.” There are strategies that will consistently give you more precise answers. The AI responds better if you give it a persona. Tell the LLM that it’s a great financial strategist, or a creative startup entrepreneur, or an AP History savant. (I often begin my prompts with “You are a brilliant editor, trained at The Atlantic and The New Yorker, who is familiar with all the rules of Strunk & White.”) This seems to give the AI a more defined context from which it draws its answers and inferences. On the flip side, it helps to give limitations in how the AI responds, such as the popular request to “explain this to me like I am fifteen” If the AI seems to be stuck in a rut of boring and predictable results, Mollick recommends telling it to “Be creative and make any assumptions that you need.”?

Another useful technique is what’s called “few-shot prompting.” Give the AI a couple of specific examples of what you want it to do and it’ll learn more quickly. When we were building a mock moderation system for a company called Speakeasy , we found the AI was much more efficient if we fed it a few specific examples of thoughtful comments and a few toxic ones.

There is another popular tactic called chain-of-thought reasoning, in which you lay out what you want the AI to do step-by-step and ask it to “show its work.” This strategy provides insight into how the AI comes up with its result, and you can determine at which step it might have gone in the wrong direction. Sometimes it can take many steps to get an AI to arrive at the right answer to a complicated logic problem, or for the AI to effectively summarize a slide deck or series of earning reports. When you arrive at a satisfactory answer, you can ask the AI: “How should I have formulated this query to receive this answer?” And it will tell you the question, like the best player on Jeopardy!

Cheers * N

I hope you enjoy this newsletter, which, thanks to your support, now has more than 450,000 readers. Please continue to forward it to anyone else who might enjoy it. They can sign up here .