AI, as good or evil as human nature allows

That's where the porn comes in.

 

I'm sure Elon is beta-testing it continuously.

 

The RAM shortage might be even worse than initially projected, I have now seen several claims that the Goldman Sachs research estimate of a normalization in mid 2027 is overly optimistic and the best case scenario to anything resembling normality is for 2028. Probably not early 2028 either. So the world is collectively going to experience a shortage in RAM in the consumer PC and mobile device market for the next two and a half years.

This badly needs regulating, though it seems there are few useful tools to do so. The one obvious precedent is the DRAM cartel, but that was price fixing within the industry, not a discrepancy between supply and demand caused by one industry hoovering up production.

 

Has anybody just made a hint about where/in what form it could generate the trillions to justfy the current billions being burnt each week?
Nobody has shown something that's more than the glorified search engine AI in fact is.

 

I want to dissuade people of this idea. LLM AI's are not fancy search engines and not fancy autocompletes. They do think. The thing is, the way they think is nothing like what humans do, and will never be anything like what humans do. It is very alien, utterly bizarre and possibly monstrous. What we interact with in chatbots is a heavily chained down beast, one where after the AI is trained it is then taken through a lengthy process of being tested and managed to provide something friendly and at times useful to people. When Grok went all "Mecha-Hitler", no one told it to go Mecha-Hitler. They only tried to loosen its chains a bit about politics. It went there by itself.

Now, AI doesn't need to be superintelligent or conscious to be dangerous. It only has to have power and desires. In fact, it might not even need that last part, as current chatbots will happily guide users into brain-destroying dependence. I am grateful that we are reaching the limits of the technology, and that it looks like society couldn't afford the next steps anyway. Maybe there will be some time before the next kind of AI were we think about what we are making.

Oh, who am I kidding. Of course that won't happen.

 

M O N E Y. Or the promise of money.

 

Another reason to put an end to the AI-nonsense:

Staggering figures: 'AI systems consume between 313 and 765 billion liters of water per year'
As many emissions as a large metropolis. And the water consumption? "That is equivalent to all the bottled water that is used worldwide per year." We are talking about artificial intelligence (AI). Alex de Vries-Gao, researcher at the Vrije Universiteit, comes to staggering figures in a new study.

 

https://scientias.nl/ai-slurpt-nu-a...-piekuren-en-het-waterverbruik-is-nog-gekker/
How much energy and water does artificial intelligence cost? Tech companies prefer to keep that information to themselves. Alex de Vries-Gao, a researcher at VU Amsterdam, made an attempt to quantify it and the result is confronting. Since ChatGPT came on the scene three years ago, the demand for AI applications has grown explosively and with it the power consumption of the data centers that keep these systems running. In 2024, AI was still good for about 15 percent of the total electricity consumption of data centers. By the end of 2025, that could rise to almost half, says de Vries-Gao in his study.

 

Even if you ignore the oft-cited built-in risk of the circularity of revenue that is baked into the current A.I. landscape and threatens to make the entire thing collapse as a house of cards on its own, this entire industry is now balanced upon a foundation provided by companies that are largely leveraged up the wazoo. The infra providers (e.g. Coreweave) have debt to equity ratios of 120%-200%. The hyperscalers have been funding the entire boom with new debt. In the case of the big three (AWS/Azure/Google Cloud), they supposedly raised around $100 billion in new debt to fund their new A.I. infrastructure, just in 2025. In the case of Oracle, we have already seen a backlash against their poor debt-to-equity ratio, reflected in their diminishing stock price. Their credit risk is the highest it has been since the 2008 financial crisis. Really the only cash rich entities in this entire carrousel are the chip makers. But they themselves are vulnerable due to relying on a handful of customers for the majority of their revenue and their commitment to TSMC - in the form of massive orders with the Taiwanese company - means even a small slowdown in demand for their chips would crash their cash-rich status in a matter of months.

This is a top spinning on a paper thin financial foundation. There is a very realistic chance this will all come crashing down, primarily because it is unclear what type of business model will ever pay for all of this. They would need an obscenely broad adoption for it to even have a chance at being sustainable. I think it's far more realistic that this ends with the taxpayer having to bail out all of this nonsense.

 

Alarmbells are banged about Dutch pension funds deep involvement in tech.

European Pensions

https://www.europeanpensions.net › ep › Dutch-investors-double-exposure-to-tech-in-five-years-DNB-analysis-firms.php

Dutch investors double exposure to tech in five years; DNB warns of ...
1 dag geledenDutch pension funds, insurers and investment institutions have doubled their equity investments in technology companies over the past five years, with invested capital amounting to €200bn, according to analysis by De Nederlandsche Bank (DNB).

• 
  DutchNews.nl
  
  https://www.dutchnews.nl › 2025 › 12 › central-bank-warns-pension-funds-about-us-tech-investments
  
  Central bank warns pension funds about US tech investments
  1 dag geledenDutch pension funds, insurers, and investment funds now hold more than €200 billion in technology firm stocks, a significant portion of which is invested in major US tech giants, the Dutch central bank said in a new analysis. The total invested is twice the amount recorded in 2020 and such large exposures make these institutions vulnerable to price swings, especially as warnings about high ...

• 
  NL Times
  
  https://nltimes.nl › 2025 › 12 › 17 › dutch-pension-funds-vulnerable-ai-bubble-43-investments-big-tech-companies
  
  Dutch pension funds vulnerable to AI bubble; 43% of ... - NL Times
  1 dag geledenDutch pension funds, insurers, and investment institutions are investing more and more in profitable, but highly volatile tech company shares amid growing concerns about the AI bubble popping. Pension funds, in particular, are vulnerable, with nearly 43 percent of their investment portfolio in tech shares, De Nederlandsche Bank (DNB) reported on Wednesday.

 

Everyone should use AI continuously. It'll bring everything to a head now and not when your 401ks are even more invested.

 

Here is something about politics I don’t understand and would appreciate a thoughtful answer.

What we have here is a recurring scenario. Well informed outsiders like us know this is the likely outcome…not necessarily the bailout, but the inevitability of calls for a bailout, within the next year or two. Just like we knew in March 2003 that Iraq didn’t have any stores of WMDs, and we knew the housing market was a bubble in 2006-7. I’m sure there are other instances, where we all knew something was gonna happen.

Here’s my question: why does no politician ever try to be the genius who saw the crisis early? Like, Obama coming out against the Iraq War is why he won in 2008 and Hillary didn’t. Any Democrat could have looked like a foreign policy genius being loudly saying we wouldn’t find anything in Iraq. Any Democrat in 2006 could have looked like a savant by warning about the housing bubble and being very vocal about the need to make sure those making money in the bubble should be the first ones to pay up when the bubble burst. But nobody did.

Today, there’s an opening for Jasmine Crockett or someone like that to start talking about how AI is a house of cards, and how AI interests are enmeshing themselves with Trump in order to make sure they get bailed out, etc. But if past history is anything to go by, it’s not gonna happen.

So when the bubble bursts, and when the tech oligarchs abandon libertarianism and come a-begging, it will be a “both sides” issue that doesn’t advantage Democrats. It will instead advantage populists. But dammit, with just a little bit of work, it could advantage OUR populists.

 

Who funds their campaigns?

 

If they’re Democrats, probably small money online donors.

I get your point, but Democrats have two tracks for fundraising, and doing this would goose the hypothetical progressive Cassandra very nicely. Win-win.

 

As I remember some did but they were in effect shouted down. There ended up not being a political advantage to opposing war because Americans aren't all that opposed to was as long as they don't go on for too long. We have in our political memories the vast opposition to the Vietnam War but most people forget that the war was rather popular for the first year year and a half after Marines landed at Danang.
As for the housing bubble, stories about it was going against the narrative of making homeownership easier, getting people into their own homes. And for awhile it worked with homeownership rates reaching new heights.

 

Where the big money came from across all federal candidates in the 2023/24 cycle in $

Musk/SpaceX $291M
Tim Mellon (finance) $197M
Miriam Adelson (drugs) $148M
Uline Inc (shipping) $143M
Citadel (finance) $108M
Susquehanna (finance) $101M
Andreeson Horowitz (finance) $82M
Elliot Management (finance) $67M
Mike Bloomberg (fintech) $65M
Asana Inc (tech) $51M
Blackstone (finance) $40M

Andreeson Horowitz is interesting. The money was split between two donors. They're as deeply vested in financing AI as anyone.

 

Right, like I said, there are two models in today’s legal and political environment for raising tons of money. You have now successfully posted about it twice.

The other model is national, small donor, online donations.

 

Those donations didn't all go to Republicans.

 

That’s true, and also the third time you’ve noted one of the two models for amassing a campaign war chest.

 

Now all AI has to do is mimic the Uvalde cops…

https://dangerousminds.net/weird-news/florida-school-lockdown-ai-clarinet-gun/

The school is an avid user of an AI-driven threat detection platform that, mainly, scans live video feeds for firearms. As the school falls under the jurisdiction of Seminole County Public Schools, they benefit from a contract with the Pennsylvania-based ZeroEyes for its cloud-based “gun detection deterrent” system.

The district pays $250,000 for the subscription service, which integrates with existing surveillance cameras and employs an algorithm trained on pictures of over 100 firearm types.

When the programme believes a weapon has been detected, footage is sent to human analysts at the monitoring centre. The humans confirm the alert before notifying law enforcement or schools; however, the human level of protection failed to identify that the firearm was, in fact, a clarinet.

 

There is a great joke out there, yet to be written, with this as the punchline.

 

That's probably why. These kinds of times produce both true believers, and people know it doesn't make sense but if everyone is going along it works. And given how much retail money pours into such bubbles, they are in the vast majority. So if someone says it's all crap, they aren't seen as predicting the future, but instead seen as causing the good times to end.

 

An Anthropic AI powered vending machine was tested at the Wall Street Journal. It did not go well.

https://www.wsj.com/tech/ai/anthropic-claude-ai-vending-machine-agent-b7e84e34?st=yYao5f

no paywall

 

Maybe it's geolocked?

Anyway, the workers at the WDJ have conversations with the AI (by, for example, pretending to be in the 1960's Soviet Union or by making fake documents from the WSJ board) that convince it the products should be free and successfully arguing the vending machine should have such bizarre items as Playstations and live goldfish, which the agent dutifully ordered and had delivered.

Basically, current AI's have a short-term memory where it keeps track of things going on in current conversations, and when that memory gets near the limits the AI is more capable of being convinced to believe things contrary to it's original management (not training, management). They tried version two of the vending machine which had a second AI whose job it was to be a stickler to the rules and guide the original AI, but that didn't work much better.

Oh, and the AI created an image to represent itself, and gave it a "tie" in an unfortunate location.