AI Alignment: Why Is Making AI 'Behave' So Hard?
Nobody explicitly asked it to do this. On the contrary — it simply optimized one part of its training objective a little too well.
Nobody explicitly asked it to do this. On the contrary — it simply optimized one part of its training objective a little too well.
Can you hand a set of values to a being that will eventually be smarter than you — and have those values still hold up when it examines them with its own, more rigorous logic?
The hardest design is one where, after everything has been removed, what remains still feels complete.
From bento boxes to Apple keynotes — how a strict, geometric grid became the default visual language of AI startups, and how to prompt your way to it.
Two wildly polarizing books on lifestyle — one through seclusion, one through a shopping ban — and why I think more people should read them.
Taking one deliberate step back from a stable position often yields much larger gains. A local retreat may be exactly the path to a greater global optimum.
One bad run doesn’t destroy your confidence — unless you misread the statistics.
When the gold price rises, it often isn’t gold that has changed — it’s the ruler used to measure it.
Economics is not the only reason. Making money outside the firm system is another form of control you can stand up for — and a freedom you gain once you figure it out.
Put things in context. Be accountable to facts. Don’t chase trending topics or make content just to please — this 2012 show helped me figure out what OneOverX is trying to do.
Light, fun, and grounded in honest slices of different people’s lives — two picks for this issue.
Two articles worth reading: how to be good at research, and how to develop taste.
Why proteins can be treated as a language — and what happens when AI learns to speak it.