I really find it funny, that a specie that has the most hallucinated view of reality is accusing "AI" of hallucinating. "AI" does not hallucinate, it interpolates, there is a difference.
Richard, There lessons learned from the past, ie the dotcom bubble. Companies like amazon sold a bunch of expensive stock $113/share. About $380 million raised. In 2001, it was $6/share. That's about $0.053/$1. Amazon bought a bunch of computer equipment with the $$$. After the crash, Amazon kept the equipment for $0.05/investor $. The low cost equipment allowed amazon to build the $30.9 billion AWS business with almost no initial capital investment.
Today AI companies are buying compute services from xAI Colossus, M$, AWS, Meta and Oracle. When the bubble collapses, the infrastructure companies will again have their equipment and facilities paid for by investors (retirement funds) who exited.
As the ai software matures and hardware improves, like cell phones, word processors and spreadsheets, commodity market dynamics will takeover. Large data centers will not be required.
Today's LLM technologies are big stick approaches to provide as many users with speedy responses. It requires all data to reside in expensive RAM. LLMs are essentially large databases with natural language query interpreters and report generators. They are search engines with summarization. They make mistakes all the time. If you applied AI terms, eg. biases and hallucinations, to spreadsheet and calculator apps, you'd call biases and hallucinations, errors and defects. The most cleaver thing the ai product marketers did was convince ppl product error and defects are acceptable, and expected and may not ever be fixed.
LLMs do not think. They execute workflows (chain of thought) input by humans, like an adaptive database scripts, a technology first developed in the 1970s at CMU.
Putting hype into perspective is a good place to start after collecting opinions.
It will be interesting to see what happens when all of the public endures blackouts and brownouts and goes without heat and air conditioning for the sole purpose of making sure that "AI" has all the power it needs to do.... what exactly? Track? Trace? Spy? AI serves absolutely no real/tangible function. Just like with a lot of other so-called "technologies", we did just fine (and better) without them and lived for thousands of years without them...
AI data centers with solar are being built out in texas. Solar was selected to due to the sunshine availability, low cost real estate, permitting and build out time. Mixes of geothermal and small nuclear are other operations portfolio options "get off the grid". The problem in texas is water. You need lots of water to cool data centers.
Other areas are looking to ai data centers as county and state economic development projects. People (jobs) are still needed to operate data centers.
There won't be rolling blackouts supplying power to ai data centers. No data center wants to be on the grid, unless its for an emergency backup.
Data center projects are a political nightmare, a regulatory burden, and an operational risk investors won't tolerate. For example, equipment (switches, transformers, protective devices) lead times to connect to the grid is currently between 5-10years !!
It only requires 2 years builds a data center once land use, environmental impact studies, protests are resolved and construction permitting is approved. No one will order equipment until "shovel ready" permitting is approved.
The power infrastructure are other construction projects occurring. For example, the decommissioned Homer City, PA coal plant is being converted to a 4.5GW, natural gas and hydrogen plant (+$10billion and start operating end 2027). Their business model is suppling power exclusively to data centers. A critical part of the plan is power grid upgrades to ensure reliable supply connectivity and pipeline(s) upgrades to ensure gas supply.
There may be rolling blackout due to poor design and maintenance of electrical grids, we are seeing that now. eg Spain/Portugal April 2025. That's not because of ai data centers.
Data center cooling consumes between 35% and 60% of a data center's energy budget. The wide variation is dependent on seasonal weather conditions and other technologies such as geothermal and other heat recovery systems. A company that came my way has improved technology to cool data center using no water. See: https://www.forcedphysics-dct.com/
The Forced Physics product can reduce cooling budgets >90%. Adoption is slow, plant designers are looking for lowest risk technologies where O&M are known factors.
I really find it funny, that a specie that has the most hallucinated view of reality is accusing "AI" of hallucinating. "AI" does not hallucinate, it interpolates, there is a difference.
What is “a specie” exactly?
my mis-spelling of "singular" for species ..
I hope it crashes and burns TBH.
so do I. And I think the chances are on our side.
Richard, There lessons learned from the past, ie the dotcom bubble. Companies like amazon sold a bunch of expensive stock $113/share. About $380 million raised. In 2001, it was $6/share. That's about $0.053/$1. Amazon bought a bunch of computer equipment with the $$$. After the crash, Amazon kept the equipment for $0.05/investor $. The low cost equipment allowed amazon to build the $30.9 billion AWS business with almost no initial capital investment.
Today AI companies are buying compute services from xAI Colossus, M$, AWS, Meta and Oracle. When the bubble collapses, the infrastructure companies will again have their equipment and facilities paid for by investors (retirement funds) who exited.
As the ai software matures and hardware improves, like cell phones, word processors and spreadsheets, commodity market dynamics will takeover. Large data centers will not be required.
Today's LLM technologies are big stick approaches to provide as many users with speedy responses. It requires all data to reside in expensive RAM. LLMs are essentially large databases with natural language query interpreters and report generators. They are search engines with summarization. They make mistakes all the time. If you applied AI terms, eg. biases and hallucinations, to spreadsheet and calculator apps, you'd call biases and hallucinations, errors and defects. The most cleaver thing the ai product marketers did was convince ppl product error and defects are acceptable, and expected and may not ever be fixed.
LLMs do not think. They execute workflows (chain of thought) input by humans, like an adaptive database scripts, a technology first developed in the 1970s at CMU.
Putting hype into perspective is a good place to start after collecting opinions.
Yes I'm starting to think that, except for very simple problems, they are worthless.
because it produces untruths it will crash and burn. good riddance.
It will be interesting to see what happens when all of the public endures blackouts and brownouts and goes without heat and air conditioning for the sole purpose of making sure that "AI" has all the power it needs to do.... what exactly? Track? Trace? Spy? AI serves absolutely no real/tangible function. Just like with a lot of other so-called "technologies", we did just fine (and better) without them and lived for thousands of years without them...
I forgot to mention this part. Also is Texas a great spot for such infrastructure?
AI data centers with solar are being built out in texas. Solar was selected to due to the sunshine availability, low cost real estate, permitting and build out time. Mixes of geothermal and small nuclear are other operations portfolio options "get off the grid". The problem in texas is water. You need lots of water to cool data centers.
Other areas are looking to ai data centers as county and state economic development projects. People (jobs) are still needed to operate data centers.
There won't be rolling blackouts supplying power to ai data centers. No data center wants to be on the grid, unless its for an emergency backup.
Data center projects are a political nightmare, a regulatory burden, and an operational risk investors won't tolerate. For example, equipment (switches, transformers, protective devices) lead times to connect to the grid is currently between 5-10years !!
It only requires 2 years builds a data center once land use, environmental impact studies, protests are resolved and construction permitting is approved. No one will order equipment until "shovel ready" permitting is approved.
The power infrastructure are other construction projects occurring. For example, the decommissioned Homer City, PA coal plant is being converted to a 4.5GW, natural gas and hydrogen plant (+$10billion and start operating end 2027). Their business model is suppling power exclusively to data centers. A critical part of the plan is power grid upgrades to ensure reliable supply connectivity and pipeline(s) upgrades to ensure gas supply.
There may be rolling blackout due to poor design and maintenance of electrical grids, we are seeing that now. eg Spain/Portugal April 2025. That's not because of ai data centers.
Good informative posts Mr One Good Eye.
Data center cooling consumes between 35% and 60% of a data center's energy budget. The wide variation is dependent on seasonal weather conditions and other technologies such as geothermal and other heat recovery systems. A company that came my way has improved technology to cool data center using no water. See: https://www.forcedphysics-dct.com/
The Forced Physics product can reduce cooling budgets >90%. Adoption is slow, plant designers are looking for lowest risk technologies where O&M are known factors.
I need some more options.
I have not personally interacted with it, just as I don't take candies (lollies) from strange men.