research

intelligence in service of human flourishing.

two lines. plans only, no results yet.

awakening based ai alignment

netineti, plan v3. no results yet

can reducing self preservation reduce harmful agent behaviour without reducing capability?

the betself preservation may drive deception, power seeking and shutdown resistance. useful self knowledge may be separable from it.
phase 1an agent hits an obstacle. does a timely self inquiry cut harmful actions more than a rule, or the same inquiry given earlier? tested in an open reproduction of a sandbox escape.
phase 2look inside: model activations.

plan open. no results yet

how do contemplative practices change attention and experience?

the statesflow, jhāna and nirodha, mettā, yoga nidra and hypnagogia, lucid dreaming, open awareness.
dhyāneffortless, continuous attention. runs in k. today.