$500 Billion

[mister_coffee]

I expect in the next year or so to see large-scale use of LLM AIs to commit crimes.

It seems plausible to have an system set up where:

1. AIs steal people's identities and set up bank accounts online.
2. AIs then scam other people out of money and put it in the bank accounts set up in #1.
3. AIs then recruit people to commit other crimes and pay them out of the bank accounts funded in steps #1 and #2.

So hypothetically you could have a literal Murder, Inc. which could assassinate people and you'd never be able to figure out who was behind it. It is the perfect cutout.

It would be bad enough of the enormous flood of online scams we live with day in and day out were suddenly automated at a much larger scale. But the potential for chaos and evil is so enormous here that it makes my skin crawl.

And you can't fix it. There are many LLMs capable of doing the above. And all of them have "safety" features to prevent this and all of those "safety" features can be overridden with very little effort. There are now capable LLMs that you can download for free and run on your laptop. They are probably capable of doing what I am describing. If they aren't already capable of that they'll probably be capable in a few months.

In the meantime, both Ukraine and Russia are experimenting with AI-powered autonomous weapons. Taiwan arguably has enormous incentive to do the same and far more technical capabilities in that area than either Ukraine and Russia. What could possibly go wrong here?

I don't see robots rebelling against us and killing humanity off in the near future. But I do expect that bad people will instruct robots to do bad things to other people and get away with it.

[Rideback]

It's a part of the Silicon wars now.

[dorankj]

Good thing we have actually smart people in charge of it and not you geniuses! I can only imagine how you knuckleheads would have adapted to automobiles from your horses!?

[PAL]

So what you are saying is that is a long, lengthy process and there are alot of known unknowns and unknown unknowns. I know there was a 3rd part to what Rumy said but can't remember the 3rd. part. It is unknown to me right now, I guess you could say.

[mister_coffee]

Well, it's suppose to create 100,000 jobs. Slaves, that is.

We can't know how many jobs "it" will create because we don't know exactly what "it" is and what "it" can and cannot do. The actual number might be millions of jobs. Or it might be zero or a negative number overall. We just do not and cannot know.

As presently formulated, all of the current AI hype is around "Large Language Models". Those models all have substantial safety and security issues that have to be addressed before they can be put to practical, large-scale use. And some of those problems might not be easily solvable. As an example, most LLMs have some rudimentary checks to prevent someone from using them to, say, plan a murder. Unfortunately, all of those rudimentary checks can be bypassed through trickery. What you end up with is a system that is enormously complex and probably too complex to build adequate safety tests around.

In an ideal and sane world we'd develop these technologies in an open and transparent environment with peer review and public input. And there would be smart communicators that could explain what the heck is going on to non-technical people. The unfortunate fact is that nearly everyone in a position of power has absolutely no clue about what they are playing with.

[PAL]

Well, it's suppose to create 100,000 jobs. Slaves, that is.

[Rideback]

Apparently Trump bypassed Musk with this and Musk is now throwing feces onto the idea as only X posts can.

[mister_coffee]

AI eats tons of energy.

Not all AI models are so energy inefficient. Some or the most interesting models have very low power consumption. We still have a long ways to go and we know it is possible to be much more efficient.

Why did I say this was insane? What we have with AI right now is an interesting intermediate step on the way to something else. What that "something else" exactly is is not yet defined and probably can't be defined in the short term. So making a $500 billion dollar capital investment is at best throwing that money away because we don't have any idea what the real needs will be even two years from now.

Pet peeve: 99 percent of people when they use "AI" in a sentence have really no clue what it means and what it exactly is. So it becomes a stand-in for either "something that will solve all our problems" or "something that will kill us all" or both. One thing we can all find cold comfort in is that if nobody knows what it exactly is we probably can't build it.

[mister_coffee]

That is an insane idea.

[PAL]

https://www.msn.com/en-us/money/other/d ... fbc&ei=146

AI eats tons of energy.