upvote
The investments in AI is in the trillion of dollars, surely that's enough capital to figure this out. It might require inventing or improving existing technology.

Before you respond "it will be too slow", the urgency here is entirely artificial – there's no reason we need to be going this fast.

reply
> It might require inventing or improving existing technology.

Or it might result in investing trillions of dollars and decades only to discover that there are only marginal gains to be had. There is no guarantee that throwing money at Problem X results in any kind of solution or that solutions to specific problems are even possible.

This is aside from the sibling commenters' very good point that this view assumes there are no concrete benefits or advancements to be made that depend on AI, which seems very unlikely at this point. Unfortunately, with or without AI, the problems we're facing socially, ecologically, medically, and scientifically are increasing. We can either use AI as a tool in the hopes we can at least delay the most serious consequences or just let them overtake us. Fossil fuels and climate change are just one of a million very pressing issues, despite the over representation it gets, and it seems highly likely that the benefits here exceed the costs. Addressing a downthread comment about labs saying we need to "slow down" - they are saying this for self-serving reasons and in their quest to ensure AI stays gatekept, not because they are worried about environmental impacts.

reply
I don't see how we could invest trillions of dollars and get nothing to show for it. Consider the Apollo and Manhattan projects, both were large investments that resulted in significant scientific and engineering progress, why would this be different?
reply
Because physics and materials science are ultimately subject to reality, they are not a slot machine you put money into and eventually get a payout.

Even your own examples tend to show that. There were in both cases rapid advances followed by diminishing returns as all low-hanging fruit was worked out. 50 years later space travel is still very expensive and difficult. Eventually we may attain reasonably economical orbital travel with sufficient advances in rocketry and materials science but there is no reason to believe any amount of money and time will, for example, give us FTL spaceships.

reply
Contrary to popular beliefs, spending money doesn’t actually grant you godlike abilities such as defying physics and chemistry.
reply
I'm no physicists, but we sure seem far from the limits, we haven't even figured out fusion yet and there are physicists, who presumably know what they are doing, working on that.

It's not like we have exhausted all avenues and are stuck. Solar panels are getting cheaper and more efficient, storage is a problem, but it has not seen that much investment. More nimble forms of nuclear fission, like SMRs, is an active area of research. As mentioned above, we are making progress on fusion.

The idea that we are anywhere close to the diminishing returns on investment in these areas seems completely wrong to me.

EDIT: The dichotomy in this discussion is striking: we need to disregard almost all negative externalities on speculative investment in AI, but speculative investment in green energy is not worth it because it might not yield sufficient results.

reply
Then why does the AI industry exist? The entire premise of this horse shit is that if we spend enough money, we can summon a demon to destroy humans
reply
> There is no guarantee that throwing money at Problem X results in any kind of solution or that solutions to specific problems are even possible.

And yet, we are currently sacrificing our entire economy, our social cohesion, most of the world’s productive resources, and the careers and futures of thousands to millions being displaced, to chase the possibility that if we computer hard enough, computer will be alive

reply
throwing large amounts of money on llms is proof that throwing large amounts of money on any problem will lead to some form of acceptable solution
reply
Explain Theranos then.
reply
> urgency here is entirely artificial – there's no reason we need to be going this fast.

What makes you believe that time is irrelevant? If AI allows for breakthroughs in mathematics, science, engineering, computing, and health care—and it does seem to be delivering on this promise so far—what in your estimation is the net cost to humankind of delaying those breakthroughs by even a year? It's not zero, is it?

reply
I mean sure, but the labs themselves are saying we need to slow down. Your position is only reasonable if the negative externalities are outweighed by the positives. Not to mention p(doom), which – unless yours is 0% – should mean we must slow down or stop immediately. We're justifying definitive downsides with speculative upsides.

More than anything though, it doesn't matter what we in tech think because we have failed to convince the general public that any of this is a good idea.

reply
Let's hope AI can solve climate change before 1.5 degrees is inevitable...

Well done humans, well done. Slow clap.

reply
> before 1.5 degrees is inevitable

You're expecting people to get working time travel from LLMs ?

reply
(that was my point)
reply
> If AI allows for breakthroughs in mathematics, science, engineering, computing, and health care.

Math, maybe. Science, where? Computing, anything measurable? Health care, where?

FYI, pre LLM techs don't count. Ain't nobody building $3tn worth of DCs for AlphaFold.

reply
Solar power doesn’t work at night, and AI data centres are a 24/7 load.
reply
I'm not sure if the car analogy applies. Data center solar power can scale without the requirement of being portable.

The problem seems to be that they want the power online quickly and they can't get solar quickly. The article mentions the inefficiency of their generators. It is costing them quite a bit extra to generate power like that, but that's their only option of they want the power immediately.

It would be nuts to do this long term. You might argue that they don't give a f about the environment , but most people will agree that they give a f about cutting costs.

reply
> Data center solar power can scale without the requirement of being portable.

You just have other issues like needing to get transmission and storage capacity built, which takes forever because of planning review, environmental assessments and NIMBYs.

reply
Not the best example, surely? My car runs on solar power (gathered from my roof)
reply
You can, but how many cars are actually powered this way? Same goes for datacenters. It's not like electrons from solar panels physically don't work with GPUs, it's just that it's hard to make it work due to other constraints.
reply
For electric cars it isn’t uncommon for people to put solar panels in their homes or their garages to charge them. I think the car themselves with solar panels won’t work because most people store the cars in garages usually, away from the sun - it is usually very harsh on everything that isn’t solar panels.
reply
> how many cars are actually powered this way?

Some. Hence "why can't they make a car run on solar power" isn't the obviously dumb question you make it out to be.

reply
1. From a pure argument lawyering point of view, my original question was "why can't [Ford/BYD] make a car run on solar power?" with comparison to a calculator, which implies a car with built-in solar panel. Otherwise the statement of "they make [thing] run on solar power" becomes meaningless. eg. "Apple made a macbook run on solar power" (because you can take a regular macbook and plug it into any solar panel.

2. There probably are non-zero amount of datacenters out there that runs off solar power. It's just not commonly practical, just like with electric cars.

3. The main point of the comment is that scaling challenges exist, and you can't take a residential use case and extend it to a datacenter. The same applies for a car, even if you could do it. A solar powered calculator is easy, just slap a tiny solar strip and it just works. A solar powered car is far harder to pull off. If you don't work from home for instance, you'll need a battery system to store the energy when you're not home.

reply
Are you intentionally trying hard not to understand?
reply
deleted
reply
This is simply an argument against datacenters. You're pointing out a negative efficiency, and the ostensible point of concentrating data processing is to produce savings.

If you can't figure out how to get renewable power to a datacenter, but you can easily figure out how to get it to the consumers of a datacenter, and you're talking about a technology that scales down to small groups well and individuals only a bit less well (but with constantly working agents utilizing compute, better than we once imagined), what's the reasoning behind datacenters at all?

If datacenters can make a case for massive increase of demand in the future, they're simultaneously making a case that individuals will have enough demand to saturate AI located closer to them or on premises. If anything, that should be a call for p2p software to help people share or borrow excess compute with and from their neighbors in exactly the same way that home solar panels use the grid.

Datacenters are also central points of failure. Distribution is resilient.

reply
Sounds like we shouldn't be building big data centers if there is only nonviable means to power them then?

I know meta and their ilk are trying to cut deal for nuclear power. I'm sure we can imagine they will take safety, the environment, etc very seriously too.

reply
>because solar panels are notoriously hard to scale

Are you high?

reply