Some interesting things to read this weekend in June

Dear Friends,

The most important story that I’ve read this month is George Packer’s dispatch from Phoenix. It’s a series of character profiles and scenes that, pulled together, show a city and a country coming apart. There’s the story of Rusty Bowers, the former Republican Speaker of the Arizona House, who prayed to God and stood tall against Trump’s election lies even as mobs came to his house. There’s Fernando Quiroz, a high-school wrestling coach who hands out bottled water to immigrants at the border. There’s Traci Page, a retiree whose well ran dry. “During this dry-up, I feel like I'm sprinting up a gravel hill and it's giving way under my feet,” she tells Packer. “I can't get ahead. And this economy, and the corruption on both sides, and the corrupt corporations coming in here—can we just catch a break?” 

The story is set in the scorching heat as the water runs out, politics grow harsher, homeless camps grow, and political activists use grievance as a grift. But, as always in Packer’s writing, there’s immense dignity to be found. At the end of his narrative, Bowers describes his dream of building a camp where people from different backgrounds can come together: “Point being, division has to be bridged in order to keep us together as a country. One at a time. That's why you get a little camp. Can I save all the starfish after a storm? No. But I can save this little starfish.”

I also loved this instant-classic Wired story about a massive stolen bicycle ring and a digital sleuth in Portland who pieces together clues—such as matching the orange shelving often shown in Facebook photos of the stolen bikes—to track down the scammers. And this story by Masha Gessen about Wahat al-Salam/Neve Shalom, a community in Israel built to foster peace between Israelis and Palestinians. Life there has not been simple since October 7th. There was, after that day of horror, one of the residents says, “this very loud silence.”

I’m a lifelong Celtics fan, for better or worse, and so I appreciated this story about the long relationship between Al Horford and Brad Stevens. I agree that Logan Airport should be renamed for Bill Russell. I love the continued quest for self-improvement of Jaylen Brown, or as his brilliant coach says, “[H]ow you have to grow, is to become vulnerable and (work) on the things that make you uncomfortable, and he does that.” And that of course reminds me of the lovely story of how Brown took on the 9-year-old national champion in chess before the NBA draft.

The most important and interesting paper I’ve read recently in AI is from Anthropic, on their attempt to understand more about the vexing question of interpretability. Can we actually understand how large language models work and make choices at a deep level, and, if we do, can we guide them in different ways? In the research, Anthropic was able to manipulate its bot, Claude, so that it was fixated on the Golden Gate Bridge. If you can nudge the LLM in that direction, could you get it to be more empathetic? Or to build a model that will teach someone chemistry but will know intrinsically that it’s a bad idea to show them how to make a chemical weapon? (The Golden Gate paper was recommended to me by an interesting source: Sam Altman, leader of a company that is a rival to Anthropic, when I interviewed him at the UN’s AI for Good summit.)

Last, it’s almost Father’s Day! Here’s Virginia Woolf on what her father meant to her: “He could be very silent, as his friends have testified. But his remarks, made suddenly in a low voice between the puffs of his pipe, were extremely effective. Sometimes with one word—but his one word was accompanied by a gesture of the hand—he would dispose of the tissue of exaggerations which his own sobriety seemed to provoke.”

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.

If you are good at math, you are likely drawn to the most complex (and most lucrative) math problem of them all: predicting the rise and fall of the financial markets. There have been many fascinating AI experiments in this area. A company in San Francisco has developed a model called TimeGPT that attempts to forecast cyclical swings in retail, housing, and the stock exchanges. Financial text mining is another popular approach: AI will analyze news articles, company reports, and regulatory filings in real-time to assess an economic sector. Last month, researchers at the University of Chicago released a paper demonstrating that a Large Language Model can be remarkably good at predicting the future earnings of a company. And this is just the public-facing research. If you’re in finance, and you figure out something that gives you an edge, you’re certainly not going to put it on Twitter. I have one friend, who will remain nameless, who says that he is funding his AI company by executing stock trades suggested by his custom AI bot.

The use of AI to analyze financial data points to a growing area of excitement in the field: the deployment of natural language processing and machine learning on specific data sets. Keeping close to the financial sector, last year Bloomberg introduced BloombergGPT, an AI-powered chatbot that has been trained on Bloomberg’s own reporting and analysis. But you can train an AI on most any set of data, large or small: people have created a chatbot that curated the collection of a museum. You can also have AI get quite personal: analyzing your sleeping habits, or, in an amazing piece of performance art, the eighty-three thousand text messages that you exchanged with your ex-boyfriend.  

It’s easy to get excited about the possibilities of chatbots trained on custom data. I’ve been trying to figure out how to train a bot on all my running logs and Strava data so that it can offer suggestions on how to train more efficiently. The advice was good, if not perfect. I also asked it for route suggestions and it had a great one—except it didn’t realize that I’d have to descend a mountain in the middle of the run. Those suggestions will only get better over time, presuming I don’t get lost somewhere deep in the Catskills beforehand.

Cheers * N