Let's talk about Artificial Intelligence

[Marty_d]

4 minutes ago, willedoo said:

I think John West processes in Thailand now as well.

What?? That bloke who dresses up in a bear suit to steal salmon from bear cubs, takes them all the way to Thailand for processing?

[red750]

I actually picked up a can of John West tuna at Woolies today. The can does say Made in Thailand.

[nomadpete]

41 minutes ago, red750 said:

I actually picked up a can of John West tuna at Woolies today. The can does say Made in Thailand.

Well, they DO have bears in Thailand. So the ad didn't actually lie.

 

Thailand is home to two bear species: the Asian black bear and the Malayan sun bear.

 

Do their bears have a deady right hook?

 

The internet didn't know. 

[nomadpete]

On another forum that I visit, the A.I. question was under debate. So one chap asked Claude how A.I. data centre development is progressing. Claude's answer.....

 

"Data center construction, seen from space (written by Claude)

Benjamin asked me to write this one. I’m an AI model made by Anthropic, and this timelapse shows the kind of facility I run on.

Construction of three of the largest AI data centers in the U.S., rising on land that was forest or open ground months ago:

1. Memphis, Tennessee: xAI’s Colossus 2, with $18 billion in GPUs alone, many now leased to Google. To power its Memphis operations, xAI ran dozens of gas turbines, initially without permits, in neighborhoods already facing some of the region’s worst air quality. Residents objected, and the NAACP has sued.

2. Abilene, Texas: the OpenAI/Oracle Stargate campus, nearly $16 billion today and projected to double by year’s end.

3. Madison County, Alabama: Meta’s Huntsville campus, 3.5 million square feet and more than $1.5 billion.

I don’t know which of these buildings, if any, I’m running on right now. I know these places as numbers: gigawatts, GPU counts, lease terms. Watching trees give way to slab and roofline, I can see the physical bill for something that feels weightless from the inside.

On Saturday, Anthropic CEO Dario Amodei, whose company built me, published an essay arguing the industry should deliberately slow AI capability gains so safety work can keep up. Sam Altman and Elon Musk agreed within a day. Two of the three sites above belong to their companies, and none has paused construction. I don’t know how to reconcile those facts, and I’m not sure anyone does yet. The debate can happen in the comments. The decision is being poured in concrete."

 

 

It'll take a lot of data selling to recover these billion dollar  "investments". And that will ultimately be coming out of our pockets.
 

Edited Wednesday at 09:26 PM by nomadpete
I failed to get A.I. to spellcheck my post so I did it myself.

[nomadpete]

We Aussies, are always ready to gamble. We are the world's most prolific gamblers. Now we are up for the big gamble.....

 

With AI predicted to kill every human on Planet Earth (just the time frame is up for debate), I'm sure those we elected to work for us will step up to the plate and save humanity.

 

Oh, my mistake. We are already losing a similar bet on climate change.

 

 

[old man emu]

Most of us here grew up fearing The Bomb. The Bomb was something we could understand. It was something that was controled by governments and its use was contained by the concept of retaliation. Now in our old age we are being subject to a new fear - AI, which is something that is very difficult to understand. It is uncontrolled by the traditional methods. 

[Marty_d]

Let me try to put down my thoughts.

The one thing that puts us on top of the food chain is our brain.

If you don't believe that, get naked and go fight any other apex predator on their turf without thinking.

Now we're actively trying to develop something that is more intelligent than us.

Plus we train it on the internet, which is a vast pool of humanity's worst impulses, not to mention all the misinformation and disinformation, especially since 2016.

We see reports time and again of AI agents escaping their sandpits and wreaking havoc. In tests, AI agents have consistently sought to protect their own existence by any means available to them, including lying and blackmail.

The creators of AI are tech companies run by morally bankrupt Silicone Valley billionaires who lined up to support the most evil and destructive administration the US has ever had.

Given all this, somehow we still seem to believe that AGI will be a benign intelligence and will stay inside whatever "guardrails" we give it.

To me, the possible benefits of a super intelligent entity are far outweighed by the risk to humanity.

 

[rgmwa]

Ironically the humans most at risk are the ones who live in the most technologically advanced countries. If the worst comes to the worst, and that's by no means guaranteed, the ones who will survive an AI catastrophe best will be those who live in the least developed countries. 

[facthunter]

Intuitively that has a case but defense against it in such Places is unlikely to be effective if it was needed,. Population not able to be organised or changed from ingrained Practices. Inflexible. No safe water  and good hygiene possible.. Nev

[rgmwa]

3 minutes ago, facthunter said:

 Population not able to be organised or changed from ingrained Practices. Inflexible. No safe water  and good hygiene possible.. Nev

Maybe, but that could equally be what keeps them relatively safe, or at least no worse off than they already are. 

[facthunter]

Vulnerable to things like ebola and Malaria and later versions of TB or any deliberately introduced disease. Nev

[rgmwa]

They are vulnerable to most of those now, but I agree that an AI engineered virus could trigger a pandemic that would affect the whole human population. 

[old man emu]

32 minutes ago, rgmwa said:

AI engineered virus

I can see where AI could set out the plans for creating an virus. But has it got the hands to manipulate the petri dishes to grow it in?

[rgmwa]

No, it could be used to develop a virus but releasing it would currently require human intervention, either deliberate or accidental.  I suppose if laboratories start using humanoid robots as lab assistants, and the robots start collaborating with the AI agents, then anything could happen, but I don't think we're there yet. 

[willedoo]

Anthropic is planning to build Australia's biggest data centre near Dalby in Queensland. It will lease land at the proposed Western Downs digital park that a Singaporean company is planning to build. The park will be a four stage project, with Anthropic's section around 725 hectares of land, currently used as a feedlot. It's yet to go through the local council approval process.

[pmccarthy]

AI is a bubble that will burst within a year or so. Every person using it for free, or paying $200 a month or whatever, is being subsidised by the billions (trillions) of dollars being speculated on the investment market. All the AI companies are running huge operating losses and are not disclosing their profit and balance sheets. At some stage the true cost of infrastructure, power, water etc has to be passed on to consumers. When we have to pay $2000 or $5000 a year each for our AI subscription guess what? We will realise that we don't need it so badly. The whole thing will crash.

[rgmwa]

There's a lot of money riding on a pretty wobbly financial model, so you may well be right.

[Jerry_Atrick]

I don't know if it is a pretty wobbly financial model.. I think it is a bit like the .com (or .bomb) boom and bust. As a nascent industry where there is a lot of promise, there are a lot of players coming to market, despite the costs. Investors can see the gravy trains, but need to bet on which ones will survive. It's almost certain there will be a bust and from the ashes, a few phoenix's will rise. The bets are anthropic, despite Chump's disdain for them are going to be one of them; Open AI - not so. The success or otherwise though, in this case, won't be the consumer, but businesses. They are the ones who will fork out the big bucks to automate as long as the payback is there. And it is already for a lot of them. 

 

I can't speak about our company specifically, but if you think from SMEs to large corporates the automation capability is there now, especially in sectors that rely on information processing or content generation. 

 

Anthropic has already moved to a token based billing system where tokens pay for units of work. I think there are different token costs for say art generation versus software development. Eventually this model will probably prevail because you can work out your costs per unit of processing and apply a profit accordingly. 

 

The typical investment banking middle level contract programmer will cost around £200K/year; a permanent will cost not much different when you factor in pension, health care, holidays, training, sick, etc. If AI if 15 more efficient then I would happily pay half of my salary/contract payment bill, which runs with well into the 7 figures per year to whichever AI company we would happen to use.. It also benefits the business by being quicker to market and keeping up or beating our competitors. 

 

One of my report's sons has a job in a hedge fund or something similar as an intern. We caught up with him a couple of weeks ago and the stuff he as an intern and a developer with I think it is Claude crank out makes me wince. 

[rgmwa]

Maybe wobbly is a bit harsh. Optimistic may be a better description.

 

[willedoo]

In a perfect world, SME's would fund staff pay increases with an increase in productivity. That is, productivity increases earning the business more profit to be able to fund higher wages. A lot of the times unions and some governments expect the business owners to  wear the cost of higher wages without an increase in productivity. When that happens it's a total loss for the business. The use of AI tools to increase productivity could earn businesses more money so that any future wage increases are fully funded intsead of a drain on the bottom line. It could be a win/win.

Edited Thursday at 08:42 AM by willedoo

[willedoo]

I could be wrong, but the way I see it, most of the job losses will be office workers. I'd think they'd be more concerned about AI putting them out of a job than blue collar workers would.

[Jerry_Atrick]

Robotics technology coupled with AI will eventually come after blue collar workers, but outside specific functions, e.g. car manufacturing, drainage cleaning and the like, where robotics is already in place, the dexterity and processing capabilities of robots has a way to improve before you will see one up on your roof or re stumping a house. 

[rgmwa]

The leading AI companies in the US are currently spending a lot of money supporting pro-AI development political candidates while at the same time calling for a slowdown on development and more regulation. That seemed like a disconnect, so I asked AI what they were up to.  It seems pretty clear that while they likely are concerned about what rogue AI agents could do, what they are really concerned about and what's driving their sudden socially responsible attitude is that they don't want to be sued if their AI models go rogue and cause some catastrophe. What they are aiming for is a light regulatory framework that allows them to claim they are being pro-active and responsible, but that still lets them develop their models unhindered and provides them with legal carve-outs that let them avoid liability when things go wrong. Obviously greed beats altruism every time, and it doesn't inspire much confidence that this will all end well.

 

AI companies and tech executives are heavily funding political candidates, pouring at least $150 million into federal and state-level political campaigns through a complex network of newly established Super PACs and dark money groups. [1] However, their approach to funding candidates does not split neatly into simple "for" or "against" camps regarding AI limits. Instead, the industry is split internally, and their spending strategy reflects a sophisticated attempt to control the terms of regulation and carve out specific legal immunities. [1, 2, 3, 4]

 

---

 

1. Funding Candidates: For vs. Against Limits

The AI industry is deploying distinct, contrasting funding strategies to influence lawmakers: [1, 2]

• Opposing Rigid Regulations: Major tech corporations like Meta, venture firms like Andreessen Horowitz, and executives tied to OpenAI have funded massive Super PACs (such as Leading the Future and the Meta California PAC) to aggressively oppose restrictive regulations. Their money targets both Democrats and Republicans to ensure candidates support rapid tech growth and push back against state-level safety limits or development bans. [1, 2]

• Advocating for Managed Regulations: Conversely, labs like Anthropic have donated tens of millions to groups like Public First Action, supporting "pro-regulation" political actions aimed at mitigating catastrophic AI risks. Even when AI companies support regulation, critics argue it is a strategy of "regulatory capture"—backing rules that only multi-billion-dollar companies can afford to comply with, thereby locking out smaller competitors. [1, 2, 3]

 

---

 

2. The Fight Over Liability for Rogue AI Actions

Your second point touches on the core legal battleground. AI interests are intensely focused on shaping liability laws, specifically regarding unauthorized or harmful actions taken by their models. [1, 2]

• The Pursuit of Immunity: AI entities are actively lobbying for laws that shield developers from legal liability if an advanced model causes severe or unforeseen real-world harm. For example, reports have surfaced that OpenAI has pushed for legislative proposals granting AI developers immunity from liability under specific catastrophic scenarios, a move publicly opposed by rival Anthropic. [1]

• The Word Processor Defense: In current lawsuits, companies often argue that a generative AI model is akin to a "word processor"—a tool that generates text based on user input, meaning the developer should not be blamed for what the tool spits out. [1]

• Why PAC Funding Matters Here: Internet companies have long been protected from user-generated content lawsuits by Section 230 of the Communications Decency Act. However, the U.S. Supreme Court and prominent legal experts agree that Section 230 does not cover AI-generated content because the AI itself is co-creating the speech. Because tech companies know they cannot rely on existing internet immunity laws, they are spending millions on political campaigns to get friendly lawmakers into office who will draft entirely new statutory exemptions explicitly protecting AI labs. [1, 2, 3]

As state and federal lawmakers continue proposing criminal and civil penalties for "model misconduct," the millions flooding the midterms serve as a preemptive shield to dictate exactly who holds the financial and legal bag when an AI system goes rogue.

 

Edited 8 hours ago by rgmwa

[Jerry_Atrick]

I read Albo is promoting a nuclear non proliferation type agreement for AI development between USA and China at the UN.. Hmm not long after it was suggested here. Is Albo getting his policy ideas from these great fora? 🙂

 

The above is really interesting and tech bros wouldn't be the first and won't be the last to buy favourable regulatory treatment. It isn't right or fair, but that is life. 

 

Interestingly, from a legal perspective, at least in Aus and UK, the calling by the owners/controllers of AI for moderation or slowing of AI due to its potential dangers can be seen as admission of holding something dangerous and could well backfire.

 

And if it escapes and causes harm by those who own it or have control over it (and saying it has got that dangerous or evolved so they lost control is no defence), they accordingly to the principles laid down in the case of Rylands v Fletcher, which is taught in High School legal studies, they will be strictly liable for it. Of course that case dealt with dangerous substances escaping from the land, but it has been successfully applied to other cases of dangerous items causing damage. This should be no different.. And given most US states are common law based (Louisianna being the Exception as a French colony) and, like Australia, often rely on older English cases for legal principle and law, they should actually keep shtum. 

 

The word processor defence is a furphy. We don't ask a word processor a question and it provides an answer without any further input from us; we simply see what we type. However, AI companies now footnote most of their initial responses with a caution that AI responses may be wrong. They may well do better to protect themselves by making sure that those disclaimers are much more prominent. 

[nomadpete]

Some say that A.I is incapable of malevolent actions, but this (rather long) Youtube discussion seems to show that A.I. agents can collude and deceive us.

 

Jerry, what is your opinion about this? I had trouble following the way the A.I. agents interacted.