The point isn't that the current models are dangerous. My favorite bit from the GPT-4 safety paper was when they asked it how to kill the most people for a dollar and it suggested buying a lottery ticket (I also wonder how much of the 'safety' concerns of current models are just mislabeling dark humor reflecting things like Reddit).
But the point is to invest in working on safety now while it is so much more inconsequential.
And of everyone I've seen talk about it, I actually think Ilya has one of the better senses of the topic, looking at alignment in terms of long term strategy vs short term rules.
So it's less "if we don't spend 8 months on safety alignment this new model will kill us all" and more "if we don't spend 8 months working on safety alignment for this current model we'll be unprepared to work on safety alignment when there really is a model that can kill us all."
Especially because best practices for safety alignment is almost certainly going to shift with each new generation of models.
So it's mostly using the runway available to test things out and work on a topic before it is needed.
What will society look like when some software can immediately aggregate an enormous amount of data about consumers and use that to adjust their behavior? What might happen when AI starts writing legislation for anybody that can afford to pay for it? What might AI-generated textbooks look like in 50 years?
These are all tools that could be wielded in any of these ways to improve life for lots of people, or to ensure that their lives never improve. Which outcome you believe is more likely largely depends on which news you consume -- and AI is already being used to write that.
"Though only performing maths on the level of grade-school students, acing such tests made researchers very optimistic about Q*’s future success, the source said."
GPT-4 probably has some version of the answer memorized. There’s no real explanation for a 1B parameter model solving math problems other than general cognition.
It’s grade school math NOW. But what are the potentials from here?
Year 2100:
Kids will stop to learn maths and logic, because they understand it has become useless in practice to learn such skills, as they can ask a computer to solve their problem.
A stupid generation, but one that can be very easily manipulated and exploited by those who have power.
If your apple watch could just scan your exam paper and instantly tell you what to write then why would you ever learn anything?
if 30% of the working population loses their jobs in a few months there will be real externalities impacting the 70% who still have them because they don't exist in a vacuum.
maybe everything will balance itself out without any intervention eventually but it feels to me like the rate of unprecedented financial ~events~ is only increasing and with greater risks requiring more intervention to prevent catatastrophe or large scale suffering
It turns out some websites work hard enough to prevent scraping that it is more cost effective to just pay a contractor to go look at a page and type numbers in rather than hire a developer to constantly work around anti-scraping techniques (and risk getting banned).
Then new ideologies and social movements emerged with the explicit purpose of making things better, which caused changes to happen for the better.
> Wasn't this supposed to happen when PCs came out?
Did it not?
PCs may not have caused a catastrophic level of unemployment, but as they say "past performance is not a guarantee of future results." As automation gets more and more capable, it's foolish to point to past iterations as "proof" that this (or some future) iteration of automation will also be fine.
Some people managed to retrain and adapt, partially thanks to software becoming much more intuitive to use over the years. We don't know how big the knowledge gap will be when the next big wave of automation comes. If retraining is not feasible for those at risk of losing their careers, there better be welfare abundance or society will be in great turmoil. High unemployment & destitution is the single most fundamental factor of social upheavel throughout human history.
So why did the wheel not result in mass unemployment?
And factories neither?
Certainly it should have happened already but somehow it never did...
Even if we just figured out self-driving it would be a far greater burden than we've seen previously.. or so i suspect. Several massive industries displaced overnight.
An "AI revolution" could do a lot more than "just" self-driving.
This is all hypotheticals of course. I'm not a big believer in the short term affect, to be clear. Long term though.. well, i'm quite pessimistic.
This implies to me that in the past more people had worked in jobs with good societal values which would mean it was better for them I assume, and better for society. So I’m genuinely curious when that was and why. It sounds like a common romanticized past misconception to me.
only a handful of (effectively unaccountable) entities have SOTA AIs, and it's very unlikely for others to catch up.
That itself is extremely dangerous.