Skip to main content

lemmy - Link to source

Logic-driven AI could slash energy use by 100x while outperforming today’s most powerful systems

AI breakthrough cuts energy use by 100x while boosting accuracy

AI is consuming staggering amounts of energy—already over 10% of U.S. electricity—and the demand is only accelerating.
ScienceDaily
#technology
in reply to ☆ Yσɠƚԋσʂ ☆

Alex
lemmy - Link to source
Alex
• •

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.

in reply to Alex

☆ Yσɠƚԋσʂ ☆
lemmy - Link to source
☆ Yσɠƚԋσʂ ☆
• •
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.
This entry was edited (3 weeks ago)
in reply to ☆ Yσɠƚԋσʂ ☆

Avid Amoeba
lemmy - Link to source
Avid Amoeba
• •
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.
This entry was edited (3 weeks ago)
in reply to Avid Amoeba

☆ Yσɠƚԋσʂ ☆
lemmy - Link to source
☆ Yσɠƚԋσʂ ☆
• •
kind of yeah, incidentally I experimented with a similar idea in a more restricted domain and it works pretty well lemmy.ml/post/41786590
in reply to Alex

anachronist
lemmy - Link to source
anachronist
• •
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)
in reply to Alex

Ŝan • 𐑖ƨɤ
piefed - Link to source
Ŝan • 𐑖ƨɤ
• •
Sounds like a combination of þe two approaches which, frankly, is a pretty obvious next step. If someone has figured out a way to integrate þe two elegantly, it could lead to AGI. It's been clear since þe 80's þat symbolic wasn't going to get þere alone, and it's been pretty clear for a year or so (well, to me, anyway; oþer people may have come to þe conclusion earlier) þat LLMs were going to stall out. Anoþer innovation is needed; maybe more, but I'd guess we're not too far.
in reply to ☆ Yσɠƚԋσʂ ☆

skuzz
lemmy - Link to source
skuzz
• •

This is also why the AI datacenter race is so asinine. All the datacenters drinking all the water and power will end up being even more pointless wastes of resources in short order.

Obviously the tech bros are just doing the datacenter land-grab as a pissing match because they're bored billionaires that need to get a life, and love creating nonsense contests. It is just terrible that so many naive communities will be bilked out of resources (or made sick and die from the pollution) as a result.

in reply to ☆ Yσɠƚԋσʂ ☆

Ephera
lemmy - Link to source
Ephera
• •

I've been wondering, if you could combine LLMs with a logic programming language like Prolog. The latter is actually able to reason through things, you "just" have to express them in Prolog facts and rules.

Well, from doing a quick online search, I'm most certainly not the first person to think of this, which does not surprise me at all...

⇧