Everyone says RL environments are the bottleneck in AI. I spent a month and $0 of compute finding the bottleneck inside the environment.
Result: harder training questions took hard-task accuracy from 19% to 59%. Stricter grading moved it near zero.
https://t.co/DWj9NS6kQY
2 big lessons I've learned lately:
1) Underpromise and overdeliver. Talking about doing the thing and doing the thing are mutually exclusive
2) Never underestimate networking. A single coffee can turn someone who wasn't hiring into someone who wants to hire you
It’s remarkable how much of life is about finding something to work on that gives you energy and surrounding yourself with people you love and then ruthlessly and militantly blocking out every single thing that gets in the way of maximizing your time spent on those two things
Any work that involves sitting in front of a laptop will be fully automated in 20 years. Full stop.
Most AI researchers at the leading labs think it will come sooner. Choose the skills you develop next very wisely.
After being in SF for only 2 weeks and talking to founders, I have realized how unbelievably easy starting a B2B SaaS is.
Building is no longer the bottleneck. Conviction in a $1B+ idea is, and a tough pill for founders to swallow is that almost nobody earns it before launch.
Met someone recently who said something I won't forget:
"I don't believe in ethics. Everyone has their own. So when you call something unethical, all you're really saying is that it breaks your rules."
Constantly reevaluating your worldview can be exhausting, but it's critical for growth. Take disconfirming information to your views with seriousness, and confirming information with skepticism.