anon1253

It always fascinated me how a lot of "state" in our neural networks is probably encoded in the extracellular fluids rather than in the neurons themselves. Biological neural networks really do not like it if you swap out their fluids (you can do this experimentally in vitro). For me it has always been one of those "nature really doesn't care about complexity" things ... it will just use whatever tools are available with little to no regard for our understanding of "boundaries". I'm not saying this complexity is strictly needed for intelligence (however that is defined) ... but I do always find it a little bit dishonest when scientists (and especially those working with pharmaceuticals) "zoom in" to the cells and then usually promptly forget that those cells are part of the much wider stage of tissues and eventually even everything outside the confines of our skin. Anyway, just my two cents.

It's a fascinating topic. I've often been surprised by its use of "tapestry" (as in "the tapestry of existence") and I never really figured out why it seems to gravitate towards such a relatively uncommon phrase. As usual I am reading too much into it (seemingly the logit for "tapestry" was higher than the others), but why? And what do these preferences tell us about the LLMs more generally? Anyway, fascinating!

RDF and semantic web used to be my goto's for this, as I believe many of the core ideas are still valid and often overlooked and sometimes even poorly re-implemented. Which says something.

However, lately I've come to like llama.cpp and friends, yes it's not ChatGTP miracle whatever but how often do you /actually/ need that? Despite its tremendous popularity, it still seems like something more people should know about. For me, I've had great fun with running LLMs locally and experiencing their different "flavors" from a more "phenomenological" (what is it like to use them) perspective rather than a technological one.