Sam Altman's Second Coming Sparks New Fears of the AI Apocalypse
wired.com
wired.com
AI does not have to be alive or truly conscious or anything to be dangerous. It just needs to be very effective at problem solving and have someone take the guardrails off and tell it to act in a self-interested way. Especially without humans in the loop.
It seems likely that the systems will continue to get much faster and more robust in terms of problem solving. It is easy to anticipate the possibility within just a few years of agent swarms being connected to pretty much everything (as directed by their human owners), and no human being able to compete. This will be a precarious position.
LLMs exist in an environment whose vulnerabke surface area is social engineering. How do you lock down a system against all possible social engineering?
Oh yah and the "system" isn't any computer system somewhere, it's the entire world itself.
So one of the problems the safetiests are trying to solve is How do you protect a messy system, the size/complexity of the entire world from social engineering? The answer is clearly not using traditional approaches which consistently fail.
And that's just one of the problems they're trying to solve.
Nuclear weapons should dispel anyone of the notion that certain “doomsday” don’t have teeth. We are quite capable of obliterating ourselves out of existence by our unsophisticated control of technical might.
Even ignoring what the technology can do, it feels me with dread that we basically have IMHO some of the worst possible people leading the charge - people who are greedy, slimy, and ruthless; who only care about money. Such a bad timeline we are in.
And who have luxury bunkers to move to in NZ or wherever for if/when it all goes wrong.
Every program written is vulnerable of being generalized into an LLM pattern.
The outputs of AI and the predictive abilities, are not scary.
The input capacity of AI will eventually eat every programmer's lunch and their programs too.
The existential risk is mapping out computer programming as an entire field and locking it away.
If you really think that the job of programmers is at risk because the current wave of AI can write any program, that also means that AI can now write a new generation of AI systems that is potentially much more powerful than the previous generation and presumably this iteration process can proceed N times until the end result is sufficiently incomprehensible for humans.
Eventually such systems will work very well and be very useful and we'll rely on it more and more.
At that point the people worried about existential risk will point out that when a vastly superior intelligence is tasked to love and protect us, a lower intelligence, it may do things for 'our own good' that we may find appealing.
I'm just about to neuter my cat today
It can read and devour every program.
Imagine a database where all possible programs are stored.
Any github project, any private repository of source code is reduced to an LLM pattern and stored away.
We can now generalize the meaning of integers and capture every single program that has been written and that you could write.
I'm not going to rewrite my comment a third time. The idea is plainly explained. LLMs can map out and store all possible programs.
Think about the structural implications of this fact. If you make money from programming, you won't like where it leads.
What AIs generate is irrelevant and non-theeatening.
What AIs are capable of storing, is a threat to programming. A "complete map".
I'm about to set my mongoose on the snakes today.
You can easily "store" all programs written in the history of humankind and all those who are yet to be written, in the same way you can "store" all books ever written by humans and those yet to be written (see "The Library of Babel" by Jorge Luis Borges)
The threat to the working programmer is having something that "understands" the programs and tells which ones are "good" (correct, useful, etc), from those who don't work (perhaps because they are a little wrong or perhaps because they are just a random jumble of instructions).
That's what I understand when I read the word "store" in the context of what you've written, but I'm pretty sure I'm misunderstanding something here.
Are you referring to the fact that the incentives behind the current wave of AI is to get them trained by copying all source code publicly available and that that is a problem because these AI models now absorbed the code without honouring the intellectual rights of the authors (and not honouring the license the authors have chosen for their creations)?
BTW, I agree that that is a problem. But it is a problem because of the outputs: the models _output_ snippets of code that is distilled from patterns (and sometimes outright copies) of the code (and other text) it was trained on. The process of extracting that information from the model *is* the process of generating an output.