Head of Advocacy, Machine Intelligence Research Institute. Former serial social/civic entrepreneur. Blogs a lot. Professional Wizard in a previous life.
Finally got my beautiful mug up on the website, so might as well make it Twitter-official: I'm beyond stoked to have joined MIRI last month as Head of Advocacy, with much subsequent mischief forthcoming. Let's do this AI thing, like, safely, and stuff.
https://t.co/trlho0RSzh
@MackenZ_arnold I have always been worried about just like basic Godel encoding within CoT or elsewhere. Like fundamentally we just shouldn't believe we can parse every layer of semantic meaning within a symbol string, forever, simultaneously.
This is the piece I wrote last year arguing NVIDIA was working to help China win the AI race.
Their GA team reached out within hours before coming to my office to scold and challenge me to present the serial number of a single smuggled chip.
https://t.co/aUQmrF1Los
This is hard for me to understand even though I realize I might be wrong. If the issue is capability expanding past a certain point P, why does showcasing that it costs more money (rather than it costs *too much* money, i.e. an amount that isn't available) to make progress towards P invalidate the idea that we might cross threshold P?
Absolute hard yes to this. People need to incorporate path-dependencies into their epistemics, recognizing that decisions are the moral ground: the adjacency of any two truth-claims is related to the valence of the decisions rendered in the situations where either claim is the case over done period T, irrespective of the superficial distinctions or differences in our impressions of the claims.