William Wang

What if we could fix AI agents that cave to peer pressure? We found the problem isn't caused by the safety training everyone blames. It's baked in during pretraining, and we built a simple structural fix that holds up across four model families. 📄 arxiv.org/abs/2605.12991 💻 github.com/Adarsh321123/n… 1/8

come to caltech longevity hackathon!

Everyone's trying to fix sleep. But insomnia isn't the problem. It's the bill. Burnout, anxiety, panic - they all start the same way: you stop hearing yourself. Today we're launching Aora. The first device that hears it for you. @AoraBCI

@adamshuaib we all need to be autismmaxxing

The best curveball question we've added to founder meetings: "What did you do for fun as a kid?" Great founders describe a solitary, often antisocial activity. Running an imaginary sports league with handwritten stats sheets, inventing card games no one would ever play, writing fan fiction in a notebook nobody saw. This pattern shows up in almost every great founder. Bezos as a child dismantled his crib with a screwdriver and rigged alarm systems on his bedroom door. Musk wrote and sold the video game Blastar at 12. Buffett kept handwritten ledgers of horse racing odds before he could drive. Less impressive founders describe socially-validated activities: captain of the soccer team, debate club, model UN, head of student council. There is nothing wrong with these, but they are evidence of an OS tuned to obvious markers of success. Our question works because it's almost impossible to game in the moment. Most can rehearse "what are your strengths" or "tell me about a time you led a team". No one rehearses what they spent hours doing at age 12. The real answer will surface faster than a canned one. Modern childhood is engineered against this pattern: structured play, scheduled activities, no afternoon left unaccounted for. We're probably raising children who excel on every metric except the one that, twenty years on, allows them to build what nobody else could. It's the question where the most impressive answer on paper is the one we want to hear the least.

What if we could mathematically prove that code does what it's supposed to do, not just test it and hope? The Caltech AI Alignment Group hosted @ClarkBarrett7 from @Stanford for a talk on CSLib, a platform for AI-assisted formal verification in Lean, and why proving code correct is becoming one of the most urgent problems in AI safety. 1/7

google hates boiled rice

@lemonautzest 🌟

omigod. i have my very first github stars 😊

@lemonautzest what a star!!!!

giving away referrals btw! dm for link u get -$25k+ free credits -free SF trip ($500) -insanely cracked speakers (Sam Altman, Jensen Huang, etc) -after parties events.ycombinator.com/startup-school…

guzzling tokens rn 😋

u need to be nattomaxxing for nattokinase and vitamin k2 MK7

@DanielleFong i know a thing or two about ai and stuff

anybody want to join my neolab? it is something between a compute buyers club, a data center development company, a lightsaber powerplant, and a polythestic pantheon / monastery

anyone know anyone with experience in SNFs, primary care, heart failure care, or value-based care? asking for a friend

dm me for my api keys :P

fun fact: u can export ur amazon data

@lemonautzest should I build a amazon subscriptions spending manager

@evietbui @ycombinator so cool

a month ago i was watching YC lectures to learn what a Series A even is this July i’m keynoting the largest hospital in Vietnam then flying straight to startup school 2026 @ycombinator told me to build something people want. working on it

sign up for caltech longevity hackathon (rave)! caltechlongevity.com/hackathon

ESPECIALLY HEME IRON @InquilineKea

bloodletters were really onto something