For the last week or so I have taken to spending about ~10 minutes per day giving direct feedback (comments, suggestions, edits) to early-career or pre-career folks interested in jumping into the AI Safety community. I’ve been doing this for a couple of reasons; one is that I’m structurally incapable of not doing it[1].
The other one is that the community is currently doing a very poor job of supporting these people, and I don’t mean monetarily (though that, too). I mean about once a day, someone posts on the BlueDot Slack or the Apart Research Discord and they have some good idea, or some brilliant idea, or some terrible idea, or just some idea, any idea, and the point is that they have gotten to the point where they are sharing the idea. This is the hard part, for them (writers).
Then comes the easy part, for us (readers): where you have to read the idea and decide in about 10 seconds if it’s worth responding to or not; typically the answer is not, and not by any fault of the people posting the ideas. It’s just that there’s a lot of them. But fortunately, as a corollary, there are also a lot of people to read them. So we don’t all have to respond to every idea.
In fact, I don’t respond to most of them, for a lot of reasons: some of them would require an essay-length response (in which case they should have maybe written an essay-length question; this is called a paper). Some of require only an emoji-sized response (👍, yes, I saw this, I hear you, you exist). Some of them might be worth a response except that the post seems very much to have been written by an LLM— which doesn’t mean that it was. Only that, because the readers in these spaces have interacted quite a lot with LLMs, they perhaps subconsciously feel while reading it that they would have a better time speaking with their own LLM than with someone else’s, and so say nothing.
But sometimes the idea is worth responding to by my own ~10-minute heuristic—not because it’s better than any of the ideas in the 3 cases above but because:
OPTION A: It’s a good idea, or a thoughtful post, or an interesting perspective, and my only job is to write down that I read it and what I thought of it; this only takes about as much time as it would take to think about the post (which I’d be doing anyway, having read it) so writing it down costs me nothing and provides them with the most valuable currency there currently is, which is attention
Alternatively, there is,
OPTION B: It’s broken and there’s some kind of fix in the idea that is easy to diagnose, takes minutes to explain, and could take them anywhere from 2 hours to 5 years to fix[2].
The nice thing is that third length of time is not my problem. It’s only the first two lengths of time which are. The problem, I think, with providing people with this kind of ten-minutes shaped feedback, is that we often feel like we are taking on the 2-to-40,000 hours required to implement our suggested feedback. We are not.
And yet, I don’t see a lot of people engaging in those ten minutes across these spaces, and certainly not habitually. There are a lot of reasons this is a bad idea (it disproportionately drives non-traditional newcomers out of the field, which are exactly the kind of people a growing field needs to reach; it prevents you from writing about and engaging with your peers’ ideas, which is an invaluable way to write about and engage with your own ideas; it turns what is nominally a community into a sort of ideas museum, which is very interesting to browse but very difficult to live inside of), but perhaps the most important one to me is: it’s mean. It’s mean to invite people into a space, tell them their ideas are welcomed and valuable and potentially the key to saving humanity from the machine, and then ignore—and I mean not individually, not you personally, not in-each-and-every-case, but structurally ignore—their smallest and easiest bid to reach out to for human feedback. It’s very mean and a lot of people don’t like to be in spaces where a lot of people are very mean.
My adult-life professional pursuits up to this point have been: teacher (undergrad TA), teacher (graduate school TA), teacher (Visiting Assistant Professor), and teacher (Assistant Professor). So in addition to being incapable of not doing it, you might also notice I enjoy doing it at under-compensated rates.
Some examples are: common and diagnosable scientific writing mistakes; a misconception about how academic conferences are run; a CV which focuses too heavily on job experience and too little on project outputs. These are real examples from my last few ~10 minutes.
For the last week or so I have taken to spending about ~10 minutes per day giving direct feedback (comments, suggestions, edits) to early-career or pre-career folks interested in jumping into the AI Safety community. I’ve been doing this for a couple of reasons; one is that I’m structurally incapable of not doing it[1].
The other one is that the community is currently doing a very poor job of supporting these people, and I don’t mean monetarily (though that, too). I mean about once a day, someone posts on the BlueDot Slack or the Apart Research Discord and they have some good idea, or some brilliant idea, or some terrible idea, or just some idea, any idea, and the point is that they have gotten to the point where they are sharing the idea. This is the hard part, for them (writers).
Then comes the easy part, for us (readers): where you have to read the idea and decide in about 10 seconds if it’s worth responding to or not; typically the answer is not, and not by any fault of the people posting the ideas. It’s just that there’s a lot of them. But fortunately, as a corollary, there are also a lot of people to read them. So we don’t all have to respond to every idea.
In fact, I don’t respond to most of them, for a lot of reasons: some of them would require an essay-length response (in which case they should have maybe written an essay-length question; this is called a paper). Some of require only an emoji-sized response (👍, yes, I saw this, I hear you, you exist). Some of them might be worth a response except that the post seems very much to have been written by an LLM— which doesn’t mean that it was. Only that, because the readers in these spaces have interacted quite a lot with LLMs, they perhaps subconsciously feel while reading it that they would have a better time speaking with their own LLM than with someone else’s, and so say nothing.
But sometimes the idea is worth responding to by my own ~10-minute heuristic—not because it’s better than any of the ideas in the 3 cases above but because:
OPTION A: It’s a good idea, or a thoughtful post, or an interesting perspective, and my only job is to write down that I read it and what I thought of it; this only takes about as much time as it would take to think about the post (which I’d be doing anyway, having read it) so writing it down costs me nothing and provides them with the most valuable currency there currently is, which is attention
Alternatively, there is,
OPTION B: It’s broken and there’s some kind of fix in the idea that is easy to diagnose, takes minutes to explain, and could take them anywhere from 2 hours to 5 years to fix[2].
The nice thing is that third length of time is not my problem. It’s only the first two lengths of time which are. The problem, I think, with providing people with this kind of ten-minutes shaped feedback, is that we often feel like we are taking on the 2-to-40,000 hours required to implement our suggested feedback. We are not.
And yet, I don’t see a lot of people engaging in those ten minutes across these spaces, and certainly not habitually. There are a lot of reasons this is a bad idea (it disproportionately drives non-traditional newcomers out of the field, which are exactly the kind of people a growing field needs to reach; it prevents you from writing about and engaging with your peers’ ideas, which is an invaluable way to write about and engage with your own ideas; it turns what is nominally a community into a sort of ideas museum, which is very interesting to browse but very difficult to live inside of), but perhaps the most important one to me is: it’s mean. It’s mean to invite people into a space, tell them their ideas are welcomed and valuable and potentially the key to saving humanity from the machine, and then ignore—and I mean not individually, not you personally, not in-each-and-every-case, but structurally ignore—their smallest and easiest bid to reach out to for human feedback. It’s very mean and a lot of people don’t like to be in spaces where a lot of people are very mean.
My adult-life professional pursuits up to this point have been: teacher (undergrad TA), teacher (graduate school TA), teacher (Visiting Assistant Professor), and teacher (Assistant Professor). So in addition to being incapable of not doing it, you might also notice I enjoy doing it at under-compensated rates.
Some examples are: common and diagnosable scientific writing mistakes; a misconception about how academic conferences are run; a CV which focuses too heavily on job experience and too little on project outputs. These are real examples from my last few ~10 minutes.