logoalt Hacker News

zarzavat • today at 1:36 PM • 2 replies • view on HN

All that evolution compresses to a GB or so of DNA (yes I know it's more complicated than that).

Humans are a bad example because of the well known width restriction leading us to be born undeveloped.

But animals can be born with varying levels of fully or partially developed neural networks despite having minimal data transfer from the parents and no apparent training data available in the womb/egg.

Nature is able to squeeze a lot more efficiency than our current AI efforts.


Replies

filoeleven • today at 2:24 PM

You're neglecting the influence of bioelectric gradients, and their storage of memory. See Mike Levin's work, where his lab created 2-headed flatworms that are genetically identical to regular 1-headed flatworms. I'm not convinced there's "minimal data transfer" when we have this whole new field opening up that shows memory is not at all limited to the brain/neural networks.

> These animals do not differ from wild-type worms in histology, expression of key polarity genes, or neoblast distribution. Instead, the altered regenerative bodyplan is stored in seemingly normal planaria via global patterns of cellular resting potential.

https://pmc.ncbi.nlm.nih.gov/articles/PMC5443973/

saagarjha • today at 1:44 PM

Seems like nature has a pretty cracked 1B parameter base model and fine tuning stack