Pinned post
Latest posts 5
in my mental model, AI that advances humanity and doesn't hurt us is to the effects of anything that reduces death (medical, crime, dangerous jobs, etc) and elevates external human suffering (poverty alleviation, housing, food scarcity, etc).
The negation of these two sets remains up for discussion of the effect AI has on humanity which is why it is important that we carefully asses how we are using AI in our day-to-day lives (what we are giving up and handing dominion to AI) and evaluate the effects it is having on our own flourishing.
The negation of these two sets remains up for discussion of the effect AI has on humanity which is why it is important that we carefully asses how we are using AI in our day-to-day lives (what we are giving up and handing dominion to AI) and evaluate the effects it is having on our own flourishing.
For people who have good harness intuition, why do you think this is? I’m genuinely curious
mini-swe-agent is 130 lines of code and it keeps on winning in harness comparisons.View quoted post on X
own your intelligence. the distribution of tasks a model is capable of doing should not be general. why pay more for a math tutor and a science tutor when you only need a math tutor? Post-train a specialized model to your task distribution (and save costs)