Coding, Data Science, A.I., Robots |

  • Thread starter Thread starter nycfan
  • Start date Start date
  • Replies: 788
  • Views: 30K
  • Off-Topic 
Oracle laid off 18% of their workforce last week. Someone tried to say "AI agents and efficiencies aren't taking the jobs, they're just reallocating their finances so they can invest more into AI."
IMO what they are saying is even worse. Cuz they're firing people so they can afford AI investments and the AI investments are likely going to mean efficiencies which are just going to get more people fired down the road.
 
One interesting thing to follow is AI data centers. Looks like they are running into supply chain constraints as well as local pushback (NIMBY). Municipalities like the prop tax revenue but they don’t provide any other benefit and lots of drawbacks.
 
Willy said:
One interesting thing to follow is AI data centers. Looks like they are running into supply chain constraints as well as local pushback (NIMBY). Municipalities like the prop tax revenue but they don’t provide any other benefit and lots of drawbacks.
Click to expand...
Biggest bottle neck is electricity followed by electrical equipment like transformers, then cooling equipment then technicians to install it.

And providers are hesitant to create new supply (especially electricity providers) as they are fearful that the whole thing will crash leaving them with excess capacity and lots of loans to payoff.

There are a lot of short term benefits like property taxes and the economic activity from new construction. But once the data center is built, they don't need too many people to keep it running.
 
gtyellowjacket said:
Biggest bottle neck is electricity followed by electrical equipment like transformers, then cooling equipment then technicians to install it.

And providers are hesitant to create new supply (especially electricity providers) as they are fearful that the whole thing will crash leaving them with excess capacity and lots of loans to payoff.

There are a lot of short term benefits like property taxes and the economic activity from new construction. But once the data center is built, they don't need too many people to keep it running.
Click to expand...
I work in commercial lending, although not specifically with data centers. That said, it’s a hot button topic because of the exposure lenders (including private) have to data centers. I think a lot of lenders are capping exposure at current levels until the path forward is clearer. All that to say, it’s not quite as simple as many believe “if you build it, AI will come.”
 
gtyellowjacket said:
Biggest bottle neck is electricity followed by electrical equipment like transformers, then cooling equipment then technicians to install it.

And providers are hesitant to create new supply (especially electricity providers) as they are fearful that the whole thing will crash leaving them with excess capacity and lots of loans to payoff.

There are a lot of short term benefits like property taxes and the economic activity from new construction. But once the data center is built, they don't need too many people to keep it running.
Click to expand...
This may be a dumb question but I believe it is your field. How is the energy required by prompts/agents/etc allocated? Is the company that “employs” the agents charged for the energy required to power the agents?
 
Willy said:
This may be a dumb question but I believe it is your field. How is the energy required by prompts/agents/etc allocated? Is the company that “employs” the agents charged for the energy required to power the agents?
Click to expand...
That's a little different part of the value chain then I typically work at but this is my guess. Sometimes there is a company that owns the data center and they would rent space inside the data center. Then the AI companies like openai and Anthropic would rent that space, put in their own servers and other equipment and pay the electricity charges either to the data center operator who would pay the utility or to the electric utility directly.

For smaller companies, they are renting server time from Amazon or Google or Microsoft and a few others. That server time includes electricity charges. Its not a separate line item. Then Amazon would pay the utility for electricity.

That may be wildly inaccurate. There may be a few other middlemen in there. I can really only speak to what I do. I buy server time from Google and I also buy AI question and responses from Anthropic. The electricity is included somewhere in those charges but not separated out.
 
Last edited:
gregh1 said:
Hallucinated citations are polluting the scientific literature. What can be done? - Tens of thousands of publications from 2025 might include invalid references generated by AI, a Nature analysis suggests.

1775489522074.png

So while journalism is dying a slow death, climate stability is dying a slow death, democracy is being tested, science is also morphing into pseudoscience? Awesome, happy Monday.
Click to expand...
This diagram is a LITERAL a description of the one and ONLY thing a LLM (Large Language Model) is capable of doing.

To whit, produce a statistically probable amalgam of stuff that it was fed in it's training dataset.

If you are expecting an LLM to do something different, the problem is not with the LLM, it's with your lack of understanding of what an LLM is capable of doing.

EDIT: You can augment an LLM with different capabilities, for example giving it access to lookup functions it knows it can call... e.g. findScholarlyPublicationCitation()... or give it access to an MCP Server which may have more domain specific knowledge, but then we're back to square one, with tailoring individual solutions for every task you need the AI agent to accomplish.
 
Last edited:
Ddseddse said:
To whit, produce a statistically probable amalgam of stuff that it was fed in it's training dataset.

If you are expecting an LLM to do something different, the problem is not with the LLM, it's with your lack of understanding of what an LLM is capable of doing.
Click to expand...
No offense, but this is not an accurate picture of what LLMs do or how they work. It is vastly misleading. The whole "it's predicting the next word" line doesn't really mean what it's often interpreted to mean.

If you want, I can: a) explain in detail; b) explain briefly; c) not explain at all.
 
You must log in or register to reply here.
Back
Top