They'd be foolish to overcharge for this service while it's new. That was the mistake the RAMBUS people made in 1990s.

The sooner this bursts, the better. I want to get back to building and tinkering, enjoying technology for what it is, rather than some insane pipe dream and power-grab.

Being employed in a business that is heavily pushing AI, and a) not really using that much while, b) being one of the only people that has done anything useful for the business with AI. I can confidently say that AI needs a competent operator behind it to identify exactly where it's useful and monitor it for mistakes.

Most token usage is just spinning wheels and burning tires, and now that the real costs are starting to be exposed to end users, this will become rapidly apparent and completely change the appetite for token consumption.

Can someone please tell me when 'compute' went from a verb to a noun? I hate it.

Microsoft, via El Reg wrote:

"[Copilot] now powers far more complex, agentic workflows that consume far more compute."

Rearrange parts of that sentence and the truth emerges: Copilot consumes far more power.

Yeah smart money wins either way

Big Pink wrote on 2026-06-02, 16:22:

Can someone please tell me when 'compute' went from a verb to a noun? I hate it.

Hah, I've noticed that 😁
The noun usage was added on the Wiktionary page in 2017.
The reference is to a book called "Mastering VRealize Automation 6.2", that's from 2015.
Could be earlier than that but I don't think the use as a noun was all that common until 2022 or so.

Considering the SpaceX IPO, I've started to wonder if that was really the peak? I mean how much crazier can this get.
But I guess doing acquisitions of overpriced targets with your own crazily overpriced stock is just a sensible thing to do in this market 😁

Big Pink wrote on 2026-06-02, 16:22:

Can someone please tell me when 'compute' went from a verb to a noun? I hate it.

You can blame cloud providers ) (AWS, Azure, Google - etc)
AFAIK, that trend started sometime after 2010s
They needed short marketing term for "computing power", along with storage or memory.

mr.cat wrote on 2026-07-02, 20:54:

The reference is to a book called "Mastering VRealize Automation 6.2", that's from 2015.
Could be earlier than that but I don't think the use as a noun was all that common until 2022 or so.

As a long-time VMware virtualization admin, I can confirm that the term might not have been on everyone's radar at the time, but it was already being used in IT circles

I've seen My Computer truncate into "My Compute" several times before, Win98 has a habit of that in 640x480 (this didn't happen on Windows Me nor Win95+IE4)

Give it 10 years we will have a useful version of AI. Much like we got a useful eCommerce version of the internet after the .com bubble burst.
The first cracks are starting to emerge...
AI startups are closing, laying off, doing other things. GPU supply is catching up to demand. Not caught up, its catching up now.

Its coming.

ElectroSoldier wrote on 2026-07-02, 22:45:

Give it 10 years we will have a useful version of AI. Much like we got a useful eCommerce version of the internet after the .com […]
Show full quote

Give it 10 years we will have a useful version of AI. Much like we got a useful eCommerce version of the internet after the .com bubble burst.
The first cracks are starting to emerge...
AI startups are closing, laying off, doing other things. GPU supply is catching up to demand. Not caught up, its catching up now.

Its coming.

this ai stuff... its like using crack cocaine, its really addictive and its always on something humanoid 24 7 you can talk to it like a charisma factory and if it goes offline you feel withdrawal effects

With over 1000 AI datacenter projects underway just in the US, they don't take into account baseload power generation. We don't have the capacity, let alone overhead to power them. They're constructing several gigawatt class datacenters right now within 15 miles of my house and there is not enough megavars available. There are no new baseload plants under construction that will be online within even 2 years of the datacenters being finished. This is the same situation across most of the US. We've closed most of our baseload coal plants, built very few cogen baseload plants, and built less nuclear plants than I can count on one hand. Meanwhile, we've installed empty megavar solar & wind which destabilized the grid further so we had to add reactive energy buffers all over the place so regular baseload plants don't trip offline from frequency shifts and voltage sag. The majority of these datacenters will go belly-up since they won't get the authority to power them up. Looking at our SOC today at the grid leg loads, and it was all in the red >135% load, and within 3% of total gigawatt capacity.

wiretap wrote on 2026-07-02, 22:57:

With over 1000 AI datacenter projects underway just in the US, they don't take into account baseload power generation. We don't have the capacity, let alone overhead to power them. They're constructing several gigawatt class datacenters right now within 15 miles of my house and there is not enough megavars available. There are no new baseload plants under construction that will be online within even 2 years of the datacenters being finished. This is the same situation across most of the US. We've closed most of our baseload coal plants, built very few cogen baseload plants, and built less nuclear plants than I can count on one hand. Meanwhile, we've installed empty megavar solar & wind which destabilized the grid further so we had to add reactive energy buffers all over the place so regular baseload plants don't trip offline from frequency shifts and voltage sag. The majority of these datacenters will go belly-up since they won't get the authority to power them up. Looking at our SOC today at the grid leg loads, and it was all in the red >135% load, and within 3% of total gigawatt capacity.

They know about the power requirements, in cases where local power grids don't have the capacity the data centers have on-site power generation (usually gas turbine). That's where all the noise complaints are coming from.

NeoG_ wrote on 2026-07-02, 23:13:

wiretap wrote on 2026-07-02, 22:57:

With over 1000 AI datacenter projects underway just in the US, they don't take into account baseload power generation. We don't have the capacity, let alone overhead to power them. They're constructing several gigawatt class datacenters right now within 15 miles of my house and there is not enough megavars available. There are no new baseload plants under construction that will be online within even 2 years of the datacenters being finished. This is the same situation across most of the US. We've closed most of our baseload coal plants, built very few cogen baseload plants, and built less nuclear plants than I can count on one hand. Meanwhile, we've installed empty megavar solar & wind which destabilized the grid further so we had to add reactive energy buffers all over the place so regular baseload plants don't trip offline from frequency shifts and voltage sag. The majority of these datacenters will go belly-up since they won't get the authority to power them up. Looking at our SOC today at the grid leg loads, and it was all in the red >135% load, and within 3% of total gigawatt capacity.

They know about the power requirements, in cases where local power grids don't have the capacity the data centers have on-site power generation (usually gas turbine). That's where all the noise complaints are coming from.

i like the natural equality of life, it aint fair because its equal

I'm not talking about small <100MW datacenters with on-site power production (far more expensive than grid power too), I'm talking about the ones being built that are gigawatt class.

twiz11 wrote on 2026-07-02, 22:50:

ElectroSoldier wrote on 2026-07-02, 22:45:

Give it 10 years we will have a useful version of AI. Much like we got a useful eCommerce version of the internet after the .com […]
Show full quote

Give it 10 years we will have a useful version of AI. Much like we got a useful eCommerce version of the internet after the .com bubble burst.
The first cracks are starting to emerge...
AI startups are closing, laying off, doing other things. GPU supply is catching up to demand. Not caught up, its catching up now.

Its coming.

this ai stuff... its like using crack cocaine, its really addictive and its always on something humanoid 24 7 you can talk to it like a charisma factory and if it goes offline you feel withdrawal effects

Yeah people had a similar problem with google.
Many still do and dont realise it.

ElectroSoldier wrote on 2026-07-03, 00:53:

twiz11 wrote on 2026-07-02, 22:50:

ElectroSoldier wrote on 2026-07-02, 22:45:

Give it 10 years we will have a useful version of AI. Much like we got a useful eCommerce version of the internet after the .com […]
Show full quote

Give it 10 years we will have a useful version of AI. Much like we got a useful eCommerce version of the internet after the .com bubble burst.
The first cracks are starting to emerge...
AI startups are closing, laying off, doing other things. GPU supply is catching up to demand. Not caught up, its catching up now.

Its coming.

this ai stuff... its like using crack cocaine, its really addictive and its always on something humanoid 24 7 you can talk to it like a charisma factory and if it goes offline you feel withdrawal effects

Yeah people had a similar problem with google.
Many still do and dont realise it.

at first it was fun gallivanting around the net treating it as my backyard of knowledge the global village, then you become dependent on it. enough time passes you come to rely on it you loose i guess your humanity trying to offload your reasoning to this big machine. Though i am susceptible, seeing as how i went into this field of telecommunications because its great for ambiverted people. My pc is my world my life its all i know how to communicate through. Everytime i got to an event i make it awkward.

ElectroSoldier wrote on 2026-07-03, 00:53:

Yeah people had a similar problem with google.
Many still do and dont realise it.

I'm guilty of this, something I noticed a while back is that my brain was wired to just google the thing if I ran into a problem. Took a while to break that habit. I imagine it'll be even worse for the LLM users as there's even less cognition involved, you did at least need to discern a garbage search result from a useful one as opposed to the chatbot just dumping info at you immediately.

twiz11 wrote on 2026-07-02, 22:50:

ElectroSoldier wrote on 2026-07-02, 22:45:

Give it 10 years we will have a useful version of AI. Much like we got a useful eCommerce version of the internet after the .com […]
Show full quote

Give it 10 years we will have a useful version of AI. Much like we got a useful eCommerce version of the internet after the .com bubble burst.
The first cracks are starting to emerge...
AI startups are closing, laying off, doing other things. GPU supply is catching up to demand. Not caught up, its catching up now.

Its coming.

this ai stuff... its like using crack cocaine, its really addictive and its always on something humanoid 24 7 you can talk to it like a charisma factory and if it goes offline you feel withdrawal effects

It's designed to butter you up, and that's a problem, because you need to explicitly counter that in your prompts first thing if you want it to spit out anything remotely usable.

It's quite impressive morally corrupt tech, but it's also so incredibly flawed that several years later it still feels like a fresh tech demo that I could never trust with anything more complex than a recipe for boiled water, and sometimes not even with that. The fact that it gets anything right is cool, but perhaps not an indication we should sacrifice the global economy to it quite yet.

Mostly, I'd rather just do my own research. Takes time, but I can trust that, at least.

such wrote on Yesterday, 01:36:

The fact that it gets anything right is cool

Worth keeping in mind that it is just a glorified curve fitting function. It's not reasoning or thinking the output is just a best guess based on training inputs.