Random Thought Thread

We don't have the energy grid for that to happen. Current LLM/AI is already stressing what's available and all the investments will stall unless they fix the power issues in the US & worldwide. That's my view, FWIW.

As someone who has spent the last decade in the energy sector, this is true. And Sam Altman recently pointed out the same thing. Without a massive effort to dramatically overhaul our nation's energy grid, we are on a path to see our AI efforts collide up against energy generation/distribution limitations sooner than later. Our permitting system, just for starters, is in desperate need of updating and streamlining. And that's just the first step to enable the rest of it to happen.

I'll stop there, to keep this from getting political. But it isn't hyperbole to look at AI development as a modern arms race, and if we don't figure out the energy component, there are going to be significant national security issues downstream.

Yeah, this keeps me up at night. Along with a number of other AI-related quandries.
 
I know that your question is rhetorical in nature and asked in jest but let's use an analogy which could perhaps underpin the gravest of all possibilities: imagine the aftermath of that atomic bomb which was dropped dropped on Hiroshima in Aug 1945 being the equivalent of the damage during the GFC back in 2008 vs the consequences of a modern nuclear ICBM being used somewhere sometimes in the future if & when that one goes BANG!
Yeah. The formula equals BIGBADABOOM
 
I am assuming that you are referring to the amount of energy needed for these data centers to grow rapidly and exponentially to service the then necessary power demand. I was referring to the magnitude of the catastrophe to the worldwide economy if the AI bubble goes BANG instead of just deflating slowly by some sort of attrition which is a better possibility than amassive explosive burst.
I think reality will set in soon about the power issues and influx of money should slow. But! Other bubbles have continued to grow despite the clear warning signs. So, I guess BANG is still on the table.
 
I hate modern fuel can and spout designs

I have no idea if it's Obama's fault but I'm going to blame him anyways
I like the Tuff Jug type and they are much better. Some have a vent whole with a little cap that eventually breaks free at the part designed to keep it on the jug while it is open, and you have to keep up with it while you’re using it. But that’s not a big deal. They also have spouts that have a long tube inside as a vent. The jugs aren’t exactly street legal for gas. They are technically “utility jugs” but apparently the are legal for DEF, as some are marked as such. IMG_9591.jpeg
 
As someone who has spent the last decade in the energy sector, this is true. And Sam Altman recently pointed out the same thing. Without a massive effort to dramatically overhaul our nation's energy grid, we are on a path to see our AI efforts collide up against energy generation/distribution limitations sooner than later. Our permitting system, just for starters, is in desperate need of updating and streamlining. And that's just the first step to enable the rest of it to happen.

I'll stop there, to keep this from getting political. But it isn't hyperbole to look at AI development as a modern arms race, and if we don't figure out the energy component, there are going to be significant national security issues downstream.

Yeah, this keeps me up at night. Along with a number of other AI-related quandries.

Don't worry, they are quietly re-introducing nuclear power .... OK for me, except for the added security issues.

Now all that's left is to demonstrate that AI can actually generate revenue ....
 
Don't worry, they are quietly re-introducing nuclear power .... OK for me, except for the added security issues.

Now all that's left is to demonstrate that AI can actually generate revenue ....

True, and I fully support it, mainly because I don't know of any other way we are going to generate that much energy. And there are advancements being made in streamlining design and production, esp. where SMRs are concerned. But the timelines involved from permit application, to bringing that baseload online to actually address the issue, are still way too long to keep pace with rapidly growing AI needs.

Waiting 5 years or more just to get a permit, then dealing with a barrage of lawsuits from environmental orgs, before you can even start building a facility that will, at best, take several more years to complete, isn't going to get us there in time. There's no good reason a permit review should take more than year, litigation needs to be constrained to a minimum if the project already meets existing code/laws, and construction can be accelerated by focusing on reproducible designs rather than treating each project as a unique one-off. This would cut the time to deployment in half.
 
True, and I fully support it, mainly because I don't know of any other way we are going to generate that much energy. And there are advancements being made in streamlining design and production, esp. where SMRs are concerned. But the timelines involved from permit application, to bringing that baseload online to actually address the issue, are still way too long to keep pace with rapidly growing AI needs.

Waiting 5 years or more just to get a permit, then dealing with a barrage of lawsuits from environmental orgs, before you can even start building a facility that will, at best, take several more years to complete, isn't going to get us there in time. There's no good reason a permit review should take more than year, litigation needs to be constrained to a minimum if the project already meets existing code/laws, and construction can be accelerated by focusing on reproducible designs rather than treating each project as a unique one-off. This would cut the time to deployment in half.
^^^ So much this! Well said.
 
So, I guess BANG is still on the table.

If you omit "ai" related hype, there was close to zero domestic growth this year, I've seen figures from 0.6 to 1.4%. DJI is up 8.4% this year, and probably something like half the alpha since 2022 has been "ai" speculation.

That's a lot of hot air.

From the trenches here, The Future(tm) of my field (software engineering) is starting to become clear. We're going to be robot babysitters, effectively more like editors. Not empowered to make design choices - the machine does that - but responsible when it doesn't work. Doing that requires experience - the machine can spam you with code all day long, but you need experienced judgement to predict how it will behave and be abused in the real world. Which means nobody wants to hire newbie programmers because they have zero experience with real-world internet-scale operational concerns.

So eventually we'll need a lot fewer of us, but that's OK, because we're not making new ones anymore.
 
If you omit "ai" related hype, there was close to zero domestic growth this year, I've seen figures from 0.6 to 1.4%. DJI is up 8.4% this year, and probably something like half the alpha since 2022 has been "ai" speculation.

That's a lot of hot air.

From the trenches here, The Future(tm) of my field (software engineering) is starting to become clear. We're going to be robot babysitters, effectively more like editors. Not empowered to make design choices - the machine does that - but responsible when it doesn't work. Doing that requires experience - the machine can spam you with code all day long, but you need experienced judgement to predict how it will behave and be abused in the real world. Which means nobody wants to hire newbie programmers because they have zero experience with real-world internet-scale operational concerns.

So eventually we'll need a lot fewer of us, but that's OK, because we're not making new ones anymore.
I have a friend who is a software programmer and, yes, things are bad out there. No jobs, unless you are AI or a visa worker. Another friend who is in DevOps and can see the writing on the wall as well. IT work is going the way machinists did with the age of CNC, only for the worse in this case.
 
Back
Top