Five essays that changed how I think about AI, and the question each one left me with.
01
What If AI Takes Over A Skill You Love?
I Have No Choice but to Bury My Talent in Yesterday · Liu Shengyu (intlsy), DeepSeek engineer
Liu wrote the attention code inside DeepSeek's latest model, and he expects AI to write it better than him within a year. He isn't scared of losing his job. He's sad about losing the work he loves. It ends with a strong opinion on who should control the most powerful AI.
What happens when we can produce more, but enjoy the process less?
Full English translation of his original WeChat post.
02
AI Could Give Us Back The Brain Space Adulting Takes Up
The Dot and the Swarm · Ethan Mollick, One Useful Thing
Mollick tests the new personal AI agents that live in your messages. One caught a wrong number in his email before it went out. Another noticed an airline credit about to expire and asked the airline for an extension.
So, what will you do with a brain that isn't constantly keeping track of unfinished tasks?
03
Your Weird Interests Might Be Preparing You For The Future
Would We Send Our Kids to HAA? · Packy McCormick & Alex Danco, Not Boring
Two founder dads ask if they'd send their kids to a16z's new build-things academy instead of college. Their take is simple. Nobody can promise a safe career anymore, so raise curious, tough kids who go deep on what they love.
Could following your interests be more practical than we were taught?
04
The Next AI Breakthrough Could Be Simulating Reality
Runway: Learning From Watching the World · Packy McCormick & Anastasis Germanidis, Not Boring
AI that learns how the world works just by watching video. The big idea is a "wind tunnel" for every slow, expensive field, so robots can practice and scientists can test ideas in simulation first.
What could we discover if testing an idea became dramatically faster?
Co-written with Runway's co-CEO, and Runway sponsors the newsletter, so this is Runway's bet told from the inside.
05
What If We Could See How AI Reaches Its Answers?
Neural Networks Are Already Clean · Nick Cammarata, Goodfire
The researcher who helped convince everyone AI is a black box now says he was wrong. In an early audit of one open model, about 90% of its neurons had one clear job. Other researchers are split, and he calls it a working hypothesis.
Imagine being able to check how AI arrived at an answer, beyond the explanation it gives you.
Short on time? Paste this into Claude or ChatGPT with the article link. You get the big idea, the best quotes, the pushback, and one question to chew on.