Donald Knuth said, “Premature optimization is the root of all evil.” AIs are built to be hardline optimizers.
Source: Structured Programming with go to Statements by Donald Knuth
Donald Knuth said, “Premature optimization is the root of all evil.” AIs are built to be hardline optimizers.
Source: Structured Programming with go to Statements by Donald Knuth
Imagine playing your first ever chess game against a grandmaster. That’s what fighting against a malicious AGI would be like.
(To Policymakers and Machine Learning Researchers)
Building a nuclear weapon is hard. Even if one manages to steal the government’s top secret plans, one still need to find a way to get uranium out of the ground, find a way to enrich it, and attach it to a missile. On the other hand, building an AI is easy. With scientific papers and open source tools, researchers are doing their utmost to disseminate their work.
It’s pretty hard to hide a uranium mine. Downloading TensorFlow takes one line of code. As AI becomes more powerful and more dangerous, greater efforts need to be taken to ensure malicious actors don’t blow up the world.
I’m not sure how you concluded that James 5:12 says you should not lie under any circumstances. I’ve always interpreted it as condemning giving oaths, especially oaths on things you have no control over.
But above all things, my brethren, swear not, neither by heaven, neither by the earth, neither by any other oath: but let your yea be yea; and your nay, nay; lest ye fall into condemnation. (James 5:12)
Did you mean to quote somewhere else?
I agree with your fundamental claim that there are lots of top tier students going to non-top schools, but I think you focused too much on SAT scores and GPA. Right now, there are so many kids getting top scores (about 5,500 students every year get a 36 on the ACT, and about 4500 students get a at least a 1570 on the SAT), test scores just aren’t enough to determine who gets in. Instead, admissions officers use a “holistic” approach, which seems rather noisy, but does factor in other real accomplishments, like getting to the IMO or starting a million dollar business.
My opinion is that we need harder standardized tests. (Maybe we on LessWrong could create one!) Until that occurs, though, I don’t think SAT scores are enough to decide that “The 25th percentile of students at University of Maryland, College Park are as good as the 75th percentile of students at Harvard”.
Why did you decide to only use rotation matrices instead of any invertible matrix? If you’re trying to find a new basis to work in, wouldn’t any invertible matrix work just as well?
You can also use Numba to speed up loops. It’s still slower than C, but it’s much better than plain Python code, and it’s really easy to implement (just import numba
and put a @numba.njit()
before your function).
I know you’re joking, but I’d like to clarify that Jesus actually said “Let he who is without sin cast the first stone,” in case some future archeologist who doesn’t know anything about 21st century religions uncovers this article. Nukes didn’t exist in the first century A.D.
I remember reading about a startup that is basically using LLMs to let you navigate through websites quicker. I’ll edit this comment if I remember what it is.
Could I point out that avoiding head injuries might not be the only reason you wouldn’t want your children to play football? You might also not want your child to adopt the culture that a lot of high school football teams have (partying, not caring about school, self-centered), which can happen quite easily if they’re around football kids 3 hours/day 6 days/week.
What Marc Andreessen has been reading. I am envious of those who get to read this many books, let alone Tyler Cowen levels of reading books. No idea how to make the time for it.
Have you considered reading Twitter less and replacing that with books?
Also by buying off or convincing those who think they have concentrated benefits that they are wrong and should stand down, as even they get more benefit from ending the diffuse costs.
This really doesn’t seem like a good way to get politics done. Is this even legal? And if it is, do you really think it makes the government better to have people effectively bribing politicians?
Have you heard about pseudoentropy? The pseudoentropy of a distribution is equal to the highest entropy among all computationally indistinguishable distributions. I think this might be similar to what you’re looking for.
No, I don’t. The resources I saw on a quick Google search were rather poor as well.
Your outline has a lot of beliefs you expect your students to walk away with, but basically zero skills. If I was one of your prospective students, this would look a lot more like cult indoctrination than a genuine course where I would learn something.
What skills do you hope your students walk away with? Do you hope that they’ll know how to avoid overfitting models? That they’ll know how to detect trojaned networks? That they’ll be able to find circuits in large language models? I’d recommend figuring this out first, and then working backwards to figure out what to teach.
Also, don’t underestimate just how smart smart 15- and 16-year-olds can be. At my high school, for example, there were at least a dozen students who knew calculus at this age, and many more who knew how to program. And this was just a relatively normal public high school.
Fact check: Mormons don’t go on missionaries until they are at least 18 for men and 19 for women.
Missionaries can be single men between the ages of 18 and 25, single women over the age of 19 or retired couples. Missionaries work with a companion of the same gender during their mission, with the exception of couples, who work with their spouse. Single men serve missions for two years and single women serve missions for 18 months.
See https://news-pg.churchofjesuschrist.org/topic/missionary-program.
Also, ever since the most recent transfer of power, Mormons have decided they want to be called “members of the Church of Jesus Christ of Latter-day Saints” instead of “Mormons”.
When referring to Church members, the terms “members of The Church of Jesus Christ of Latter-day Saints,” “Latter-day Saints,” “members of the Church of Jesus Christ” and “members of the restored Church of Jesus Christ” are preferred. We ask that the term “Mormons” and “LDS” not be used.
See https://newsroom.churchofjesuschrist.org/style-guide.
Also, could I seriously advise not mimicking the Mormon missionary program? Mormon missionaries are basically cut off from everyone and everything except the Mormon church. Until about three years ago, they weren’t even allowed to call home more than twice a year. Apparently it’s also so stressful that about half of them return home early, where they’re further shamed for not meeting the exacting expectations of their church. It’s basically human trafficking in the name of religion. You can read all kinds of mission horror stories on (the admittedly terribly biased) https://www.reddit.com/r/exmormon.
Is this serious? I find it somewhat ironic that your deontology is completely closed-minded on its belief about narrow-mindedness.
I thought masters’ theses were supposed to be about new research (and maybe bachelor theses too?). Is this not the case?
Have you experimented with subtracting from the loss? It seems to me that doing so would get rid of the second term and allow the model to learn the correct vectors from the beginning.
Humans have biases they don’t even realize. How can we verify an AI lacks such biases?