I'm writing a book about epistemology. It's about The Problem of the Criterion, why it's important, and what it has to tell us about how we approach knowing the truth.
I've also written a lot about AI safety. Some of the more interesting stuff can be found at the site of my currently-dormant AI safety org, PAISRI.
Posts
Wiki Contributions
Comments
READ BLOGS? GET AN RSS READER
My personal recommendation is a service called Feedrabbit. It's simple, cheap, and sends RSS to your email, which makes it easy for me to process it alongside all my other sources of notifications.
As a side note, I honestly wish I could get more things into my email so I could process them all in one place. I realize this isn't how most people relate to email, but because I have my inbox well managed and maintain an inbox zero system, it makes email the ideal place for me to consume new things (incidentally, this is one of the things I really like about Substack: it sends me email!).
According to METR, the organization that audited OpenAI, a dozen tasks indicate ARA capabilities.
Small comment, but @Beth Barnes of METR posted on Less Wrong just yesterday to say "We should not be considered to have ‘audited’ GPT-4 or Claude".
This doesn't appear to be a load-bearing point in your post, but would still be good to update the language to be more precise.
1When you take LSD, it's necessary to have a sober trustworthy person around so you don't think "cars aren't real" and go wandering into traffic. The same goes for mind-altering meditation with similar effects. If I had common sense, I would have kept going to the Zendo. That way I'd have been around kind, experienced people who could remind me that cars are real. Instead, I thought to myself, I don't need teachers. I've taught myself lots of things before. I can traverse this territory just fine myself.
I'm sorry you had to go through a lot to realize what you already realized but didn't believe at first.
This is one of the big drums I keep beating with rationalists and EAs and other sorts of nerds who want to get into meditation. You might think you can do it alone, and some of you can, but even if you can it's not fully practicing the dharma.
This seems to be a trait common in almost all Buddhist traditions, going back to the earliest stories where the Buddha tells Ananda that good spiritual friendship is the whole path. And even in Zen, which literally starts with dudes sitting alone in caves, treats such solitary practice as something to be done temporarily by experienced practitioners.
I'm glad you came through it okay, and it certainly seems like it was ultimately a necessary if painful part of the path for you.
Worth noting that English does try to get at these distinctions, but often by using phrases or context:
- motherly love, fatherly love, brotherly love, sisterly love, etc.
- "I love [X]" where X is clearly a non-person so there's no real confusion in context
- "I love them" vs. "I'm in love with them"
- "God/Jesus/etc. loves me" is clear to people who feel this that it's by analogy to the simple feeling of love, like one feels loved by one's parent when a small child
I basically reject the premise of the OP that there's any real confusion when "love" is read on context (except in cases of intentional ambiguity). It's only lacking context that "love" seems like it's a confused concept.
This is anecdotal, but it's also the primary reason I'm not worried.
A friend of a friend heard about the phytoestrogen thing, and wanted to transition from male to female, so attempted DIY hormone therapy by eating a large quantity of soy products, as in, several ounces of soy at every meal, drinking the liquid they pack tofu in, drinking lots of soy milk, etc.
There was no noticeable effect after months of trying.
Obvious caveats on this story for n=1, but additionally given the high prevalence of soy in the world, my read is that either even a little soy is a problem and we don't see clear effects of eating more soy because it's already had most of its impact on us or it's just not that big a concern. Either way, unclear that anything short of extreme action to avoid soy would be likely to show an effect if there is one, and then we'd have to worry about confounders from whatever those extreme actions are.
Feels like this has too much wiggle room. Like what counts as an "easy" problem of consciousness and what counts as "transcending" it? Generally good definitions avoid words that either do too much work or invite judgement calls about what counts.
It really helps if we just taboo the word "consciousness" because people have too many implicit associations wrapped up in what they want that word to mean.
On a day to day level, we want "conscious" be to a stand-in for something like "things that have subjective experiences like mine". This is unfortunately not very useful, as the world is not carved up into thing that are like this and not, other than for other humans.
On the other, if we try to get technical about what we mean for things to be conscious, we either end up at panpsychism by deflating the notion of consciousness (I'm personally supportive of this and think in many cases we should use "consciousness" to refer to negative-feedback control systems because these are the smallest unit of organization that has subjective information), or we end up with convoluted definitions of consciousness to add on enough qualifiers to avoid deflation.
"Consciousness" is a word people are really confused about and have lots of different competing intuitions about what it should mean and I really wish we'd just stop saying it and talk about what we mean directly instead.
Much of this depends on what kind of AI we get and how long we live in relatively the same conditions as that AI.
The unstated assumptions here seem to me to be something like:
- AI provides relatively fixed levels of automation, getting gradually better over time
- AI doesn't accelerate us towards some kind of singularity so that society has time to adapt to tiering
I'm pretty suspicious of accepting the second assumption here, as I think just the opposite is more likely. But, given the assumptions Acemoglu seems to be making, a two-tiered society split seems a likely outcome to me.
Sort of a tangent, but when I ride I technically fall into the "strong & fearless" group, but don't feel like it.
I'd much prefer protected bike infrastructure, but for a variety of reasons it's often unavailable. In those cases, at least in an urban setting, I generally prefer to ride in the lane with cars over using an unprotected bike lane. To me this is obvious:
- in the lane the cars will definitely see you
- if you ride far enough into the lane cars can't pass you by "squeezing by", getting dangerously close
- less likely to put yourself in precarious situations at intersections
Unprotected bike lanes seem like the worse possible option, and find it strange that anyone falls into the "enthused & confident" category.
NB: At the moment I forget where I have this idea in my head from, so I'll try to reexplain it from scratch.
There's a well known idea that external behavior can only imply internal structure if you make assumptions about what the internal structure is like. This is kind of obvious when you think about it, but just to make the point clear, suppose we have a black box with a red button on top of it and a wire coming out the side. When you press the button, the wire carries a current, and then it goes dead when the button is not pressed.
You might be tempted to say it's just a switch and the button is closing a circuit that transmits electricity on the wire, but that's not the only option. It could be that when you press the button a radio signal is sent to a nearby receiver, that receiver is hooked up to another circuit that does some computations, and then it sends back a signal to a receiver in the box that tells it to close the circuit by powering a solenoid switch.
You can't tell these two cases apart, or in fact infinitely many other cases, just from the observed external behavior of the system. If you want to claim that it's "just a switch" then you have to do something like assume it's implemented in the simplest way possible with no extra stuff going on.
This point seems super relevant for AI because much of the concern is that there's hidden internal structure that doesn't reveal itself in observable behavior (the risk of a sharp left turn). So to the extent this line of research seems useful, it's worth keeping in mind that by itself it will necessary have a huge blind spot with regards to how a system works.