Hacker News Reader

Comment by thelastgallon | Hacker News Reader

"Although it will happen incrementally, astounding triumphs – fixing the climate, establishing a space colony, and the discovery of all of physics – will eventually become commonplace. With nearly-limitless intelligence and abundant energy – the ability to generate great ideas, and the ability to make them happen – we can do quite a lot." - Sam Altman, https://ia.samaltman.com/

Reality: AI needs unheard amounts of energy. This will make climate significantly worse.

View on HN

> AI needs unheard amounts of energy…

… and it always will? It seems terribly limiting to stop exploring the potential of this technology because it’s not perfect right now. Energy consumption of AI models does not feel like an unsolvable problem, just a difficult one.

AI, i.e. deep learning, consumes huge amounts of energy because it has to train on gigantic amounts of data and the training algorithms are borderline efficient. So to stop using huge amounts of energy someone would have to come up with a way to get the same results without relying on deep learning.

In other words all we need is a new technology revolution like the deep learning revolution, except one centered around a radically new approach that overcomes every limitation of deep learning.

Now: how likely do you think this is to happen any time soon? Note that the industry and most of academia have bet the bank on deep learning partly because they think that prospect is extremely unlikely.

This feels like a “fixed mindset” about energy, i.e. the way we generate and allocate energy today can not change. Of all the “wasteful” things we do with energy AI is not the top of my (very subjective) list of things to shut down.

No what I'm saying is that for the way we use energy to train neural nets can change in principle but nobody knows how to do it.

I think you should study the tragedy of the commons to understand why it always will.

Energy isnt a common. It costs.

Also the tragedy of the commons is based on a number of flawed assumptions on how commons work.

Wow that quote is absolutely crazy. “Discovery of all of physics”. I think that’s the worst puffery/lie I’ve ever heard by a CEO. Science requires experiments so the next LLM will have to be able to design CERN+++ level experiments down to the smallest detail. But that’s not even the hard part, the hard part is the actual energy requirements, which are literally astronomical. So it’s going to either have to discover a new method of energy generation along the way or something else crazy. The true barrier for physics is energy right now. But that’s just the next level of physics, that’s not ALL of it.

Edit: Also why are you getting downvoted...

I can't imagine anyone writing that paragraph of his with a straight face. The chuckling must've lasted a week.

So far it seems that AIs appetite for energy might finally pushing western countries back to nuclear, which would make climate significantly better.

A world where we produce N watt-hours of energy without nuclear plants and a world where we produce N+K watt-hours of energy with K watt-hours coming from nuclear has exactly the same effect on the climate.

Unfortunately no, this is not how it works.

The relative quantity of power provided by nuclear (or renewables, for that matter) is NOT our current problem. The problem is the absolute quantity of power that is provided by fossil fuels. If that number does not decrease, then it does not matter how much nuclear or renewables you bring online. And nuclear is not cheaper than fossil fuels (even if you remove all regulation, and even if you build them at scale), so it won't economically incentivize taking fossil fuel plants offline.

Nuclear cannot "make the climate better" but can perhaps slow the destruction down, only if it replaces fossil fuels, not if it is added on top due to increased energy consumption. In that case it's at best neutral.

Nuclear and everything we were using before is probably not better than just everything we were using before. Hopefully we can continue to reduce high emission or otherwise damaging power production even while power requirements grow.

Power means making things and providing services that people want, that make their lives better. Which is a good thing. We need more power, not less.

Almost all power consumed ends up as thermal energy in the environment.

That would be fine if we weren't preventing it from radiating out with greenhouse gases.

I guess that global warming is a good thing, then

Only if they build the reactors then go bankrupt leaving the rectors around for everyone else. Likewise if they build renewables to power the data centres.

Almost all energy consumed for computation turns into heat. Increasing energy consumption therefore doesn’t help the climate, in particular if you don’t source it from incoming heat from the sun (photovoltaics or wind energy).

Reply on news.ycombinator.com