As an intuition pump, imagine Neuman was alive today; would it be worthwhile to pay him to look into alignment? (He explicitly did contract work at extraordinary rates IIRC). I suspect that it would be worth it, despite the uncertainties. If you agree, then it does seem worthwhile to try to figure out who is the closest to being a modern Neumann and paying them to look into alignment.
Tomas B.
I started writing this for the EA Forum Creative Writing Contest but missed the deadline. I posted this on LessWrong but figured I would post it here, too.
The Maker of MIND
In general, I have noticed a pattern where people are dismissive of recursive self improvement. To the extent people are still believing this, I would like to suggest this is a cached thought that needs to be refreshed.
When it seemed like models with a chance of understanding code or mathematics were a long ways off—which it did (checks notes) two years ago, this may have seemed sane. I don’t think it seems sane anymore.
What would it look like to be on the precipice of a criticality threshold? I think it looks like increasingly capable models making large strides in coding and mathematics. I think it looks like feeding all of human scientific output into large language models. I think it looks a world where a bunch of corporations are throwing hundreds of millions of dollars into coding models and are now in the process of doing the obvious things that are obvious to everyone.
There’s a garbage article going around with rumors of GPT-4, which appears to be mostly wrong. But from slightly-more reliable rumors, I’ve heard it’s amazing and they’re picking the low-hanging data set optimization fruits.
The threshold for criticality, in my opinion, requires a model capable of understanding the code that produced it as well as a certain amount of scientific intuition and common sense. This no longer seems very far away to me.
But then, I’m no ML expert.
Current scaling “laws” are not laws of nature. And there are already worrying signs that things like dataset optimization/pruning, curriculum learning and synthetic data might well break them—It seems likely to me that LLMs will be useful in all three. I would still be worried even if LLMs prove useless in enhancing architecture search.
This comment is pretty stupid.
>I suspect that this doesn’t work as an idea, largely because of what motivates mathematicians at that level.
How confident of this are you? How many mathematicians have been offered, say, $10M for a year of work and turned it down?