AI is consuming staggering amounts of energy—already over 10% of U.S. electricity—and the demand is only accelerating. Now, researchers have unveiled a radically more efficient approach that could slash AI energy use by up to 100× while actually improving accuracy. By combining neural networks with human-like symbolic reasoning, their system helps robots think more logically instead of relying on brute-force trial and error.
LLMs have some interesting properties and certainly can do a good job sifting through large amounts of raw data. They are however a very brute force approach compared to say a network routing protocol. Sooner or later people will start to realise (again) that engineering is about trade offs and you need to work out what your constraints are and stop trying to solve every problem with massive amounts of multiplication.
Basically, the idea is to use a symbolic logic engine within a dynamic context created by the LLM. Traditionally, the problem with symbolic AI has been with creating the ontologies. You obviously can’t have a comprehensive ontology of the world because it’s inherently context dependent, and you have an infinite number of ways you can contextualize things. What neurosymbolics does is use LLMs for what they are good at, which is classifying noisy data from the outside world, and building a dynamic context. Once that’s done, it’s perfectly possible to use a logic engine to solve problems within that context.
So for peasants running Chairman Xi’s LLMs on local GPUs, we could try the largest model we can run and have it generate scripts to run instead of having the model do the actual processing of bulk data, to get more out of it.
Full circle. After a big orgy of trying to make ever larger word guessing engines write software we rediscover that computers are fundementally logic machines (and also word guessers were never intelligent)
So algorithms then?
LLMs have some interesting properties and certainly can do a good job sifting through large amounts of raw data. They are however a very brute force approach compared to say a network routing protocol. Sooner or later people will start to realise (again) that engineering is about trade offs and you need to work out what your constraints are and stop trying to solve every problem with massive amounts of multiplication.
Basically, the idea is to use a symbolic logic engine within a dynamic context created by the LLM. Traditionally, the problem with symbolic AI has been with creating the ontologies. You obviously can’t have a comprehensive ontology of the world because it’s inherently context dependent, and you have an infinite number of ways you can contextualize things. What neurosymbolics does is use LLMs for what they are good at, which is classifying noisy data from the outside world, and building a dynamic context. Once that’s done, it’s perfectly possible to use a logic engine to solve problems within that context.
So for peasants running Chairman Xi’s LLMs on local GPUs, we could try the largest model we can run and have it generate scripts to run instead of having the model do the actual processing of bulk data, to get more out of it.
kind of yeah, incidentally I experimented with a similar idea in a more restricted domain and it works pretty well https://lemmy.ml/post/41786590
Full circle. After a big orgy of trying to make ever larger word guessing engines write software we rediscover that computers are fundementally logic machines (and also word guessers were never intelligent)