>>38312
>I personally believe that any non return path, no permeable material, motor/actuator will always be inferior to the one with a return path.
I don't have sufficient training in EM/Field-theory
(yet) to know any better one way or other, but the
instincts of my mind's imagination tell me this is the correct take.
>why does every single electric car makers motors have permeable material to concentrate the magnetic field?
We have a video linked here on the board somewhere about the remarkable stator, etc., design of the newer Tesla motors. It would be nice to know more about the manufacturing techniques used to produce them! There's also another video about the 'print-in-place' method used behind an even more-sophisticated, smol'r design motor (in response to your 'make coils:no winding' comment posted above).
<--->
>"1.5 million parameters is all that is needed to capture the sophistication of the human body for humanoid robot training."
<sauce: bull's a*rse
With all due respect to those esteemed researchers, I think this is
yet to be determined! :D
We simply can't say something like this definitively yet, I deem. But what we
can say definitively is that the eminent
Carver Mead has done extensive research on so-called
Neuromorphics (being the founder of the field back in the day) [1]. And what Dr. Mead, et al, says is that the "intelligence" of creatures gets 'pushed out' to the periphery of their body structures. In other words, for us humans, a very large part of our own sensorimotor / kinesthetic "understanding" happens not within our brains --
but within our peripheral spinal-column + limbs neural tissue!
Certainly such a model of reality seems to me (at least with merely a surface, cursory view on these matters) to warrant
against using just a simple, "LLM-like" discourse as the correct method to approach these problemspaces with. /$0.02 .
<--->
As to your back-of-the-napkin estimate of the h/w power needed for our robowaifus to be safe & successful, effective & pleasing waifus... again, hard to say yet. OTOH, I'm rather-confident we can, in fact, solve the basic kinematic needs with a reasonably modest array of smol SBCs & MCUs + sensors, etc. [2] Its the very
language/emotions &
world-modelling/judgement problems themselves that may need an inordinate amount of compute to manage well, I think. Stay tuned.
I think most of us plan to use some sort of home server setup for offloading robowaifu compute. I personally mean rather to have it all onboard in the nominal case, and will be satisfied (at least at first) with having conversations not much more extensive than her saying
"Chii?",
"Chii!",
"Hideki?",
"Hideki!!" After all, teaching her about life & things is
where all the fun is! :DD
Cheers, Grommet. :^)
---
1. (cf.
>>12828,
>>12857,
>>12861 )
2. with the fundamental proviso that the software itself
running on said hardware is highly-efficient, realtime-compatible, C++ & C code!
Edited last time by Chobitsu on 05/09/2025 (Fri) 09:09:07.