Why Gradient Descent Still Fails: Key Preconditions Identified
On a first read, this paper seems far ahead of the pack in terms of (1) understanding some reasons why a task might stay difficult even in the face of gradient descent, and (2) distilling out propositions they'd need to somehow verify before they started expecting nice things.
Superintelligent AI Threatens Everyone, yet Hope Persists
Machine superintelligence would extinguish Democrats, Republicans, British, Chinese, scientists, cab drivers, and polar bears. It is a sign of hope that all of those now seem to be saying they'd prefer otherwise (except the polar bears). https://t.co/4fAKRAslD5
Ban AI Human Impersonation to Test Global Regulation
There are nearly no good reasons for an AI to ever impersonate a human. Making that illegal in as many countries as possible, and improving state capacity to enforce it, would be a good trial balloon for human civilization's ability to...
MIRI Technical Governance Fellowship Applications Due Feb 22
Apply for MIRI Technical Governance fellowship program; deadline now set to Sun, Feb 22nd. https://t.co/taJGPK1Wla
Prioritize Lightcone: It Needs More Immediate Funding
MIRI and Lightcone's SFF-matched fundraisers are coming down to the wire the next 3 days (as such often do). MIRI has $1M of $1.6M remaining on its match. Lightcone needs $1.6M of $2.6M. **If** you're equally motivated to donate to...