Large Language Models to Small Physics Models

Riffing on AI architecture with the (boys?)

"" If you want logic, look at 1990s automated theorem provers and descendents. Nothing new "" 

"" LLMs doing bumblefuck reinvention of logic from emergent grammar is just a waste of energy "" 

"" Btw, logic is a canonical abstraction over physics. The only thing better than an ATP for logic, will be whenever they transition LLMs into LP(hysics)Ms. Or SPMs for that matter.

One of the quick napkin calculations I ran a few months back before taking a break from computational work was ... to figure out how many voxels fit in a number of bytes. E.g. you can use abstract data structures such as spheres or cubes or tetrahedrons wtv. For ease, consider cubes. A 2-cube is 8 bits. A 4-cube is 8 bytes ... and that's good enough for roughly guessing the shape of a thing already. A 16-cube is only 0.5 KB ... but you can describe most common household items with it.

Then we start to refer back to the sorts of chip hardware being used for popular software today : they're matrix math ops used to run markov models, but they will eventually be repurposed to run spatial ops. It goes full circle back to what game chips were originally made to do. LLMs are just a distraction because neuromorphic architecture helps with language games. But languages are just abstract games over underlying reality, whose data is ~4D geometry. We will see more of such software, soon, in anthropomorphic agents. ""

--

Ranting about glorified reinventions of logic from bumblefuck emergent grammars on socmed. 

"" It's simply a crude model of logic, which costs oodles of resources more to generate via emergence from chaotic systems instead of just using NAND gates. This entire approach needs to eat a bag of dicks and move onto proper physical modelling asap i am so tired already 😂"" 

"" ( replying to : [ Logic aside, do you think psychological modeling has no value? ] )

for starters the entire formal discipline of psychology is FAR from being mapped precisely to neurology 

But, let me step back and accept that you probably mean lay-psychology, or natural language psychology, i.e. the notion of human minds as the man on the street appreciates them. Sure, it is useful to model all sorts of romantic familiar behaviour. But that just puts such efforts in the same spot as formal psychology above. 

The only good psychology is ever going to come from physical models. Meanwhile what we have is critical theory with physics envy.""