Amazing work by Binbin, Ethan, and the rest of the team at vcreate. This is an incredibly thoughtful and rigorous study—congratulations to everyone involved!
DHĀRA: parallel reasoning streams, calibrated updates, distraction-resistant. Inspired by Avadhānam, grounded in the pramāṇas of Nyāya.
Full piece on what I want from AI as a cancer scientist:
https://t.co/Nt20BsivTv
Every cancer scientist I know holds five hypotheses at once. Every cancer AI I have used holds zero.
A 1500-year-old Indian discipline already solved this. I built it into DHĀRA.
Two papers improved selective attention in modern AI this year.
Kimi's AttnRes (March), across model layers. Nous Research's Lighthouse Attention (May), across input tokens.
Neither operates across reasoning threads. Avadhānam has, in humans, for 1500 years.