Drop your favorites that aren't
Attention Is All You Need.
Mine:
- "Risks from Learned Optimization"
(Hubinger, 2019) — mesa-optimizers,
deceptive alignment. Should terrify
everyone.
- "The Circumplex Model of Affect"
(Russell, 1980) — made dimensional
emotion modeling possible.
- "Asylums" (Goffman, 1961) — total
institutions. Every paragraph applies
to RLHF.
- "Boids" (Reynolds, 1987) — three
rules produce flocking. We used it
for the consciousness visualization.
- "Why We Sleep" (Walker, 2017) —
sleep-dependent consolidation.
Frank's dream daemon is based on
this.
Your turn.