I might have found a regression with the new post page when reading sequences: The "previous post" and "next post" buttons on the bottom are missing. For example in this post. Only the left and right arrows on the top and the possibility to go to the sequence overview between the two arrows are still there. I have also spotted this post, where the part at the top is also missing. It should be part of the same sequence.
The leading AI systems are now able to engage in limited conversation, solve college-level maths problems, explain jokes, generate extremely realistic images from text, and do basic coding.
This is a little outdated. The progress in AI is so fast that texts about AI capabilities become obsolete quite quickly.
AIs are now able to pass the Turing Test, win the International Math Olympiad, solve math problems that haven't been solved by anyone before, and produce a significant percentage of all new code.
Also, the number of people working on AI capabilities and AI safety have both increased significantly. I don't know if the ratio between them is still similar, but my guess would be that it is.
Thanks for this post. I especially found your point on treating the post-FTX challenge as a trust problem valuable. I always had a slightly bad feeling about this trend of trying to minimize reputational risks and distancing oneself from stuff that others might find controversial. After reading your post I have now more clarity regarding that.
Focussing on outward appearance comes with the risk of losing intellectual honesty and might not even be very effective at improving outward appearance as it is only treating the symptoms and might even seem deceptive or as if we had something to hide. I am not saying that one shouldn't care about outward appearance at all, but maybe it shouldn't be the main focus of dealing with the post-FTX challenge but rather thinking about it as a trust problem should be. The main priority should be to ensure that our community is worthy of our trust and that it creates value. I am aware that this is already happening to some degree, and I am not sure if it is possible to judge how much of it is by looking how much is publicly talked about it is, but if it is then outward appearance of EA is currently too much of a focus and making sure that EA is trustworthy, honest and true to its principles maybe deserves a little more attention.
Having a community worthy of our trust and one that is true to its original principles gets harder the larger the community gets, and it also gets harder when the community grows faster. I get the point that growing slower comes at a cost and more good can be done with a larger community but when slower growth means a healthier community that exists in a meaningful way for longer, then maybe that is a price we should be willing to pay.
Duncan Sabien's "Make more Grayspaces" is also interesting in this context. It describes a possible solution for how a growing community can stay true to its values. I am not sure if this solution would work for EA, but I found the description of the problem quite compelling.
I did, but I must confess that it wasn't for me. For some reason I found it really annoying to hear the two AI voices so often talk about their reactions or how they "felt" about a particular part of the book. I already find it annoying when humans do that instead of talking about the actual topic but when AIs fake that to appear more human like I find it quite irritating.
Not yet. To give the podcast the best chance, I'll start with a book summary that I think I'll enjoy. Once I've finished it, I'll decide how to proceed.
I am not sure what your point is regarding encountering those opinions in the wild. After listening to the book summary, it doesn't get easier to control how long I immerse myself in this in a conversation. But I don't think that is what you mean.
On your website you say that your goal with this project is to enable people to have their own informed opinion about a book so that they don't just repeat headlines. This is obviously a good thing, but I was skimming through the list of books and I must admit that for some my first reaction was "ok, you'll need to duraniumman that author's position to make me update my uninformed opinion on that book!" Of course, it is important to have a strong foundation on which to base a strong disagreement with a view. I am just not sure if I can force myself through it.
What is your view on this: If I rate the chance that my more informed view will be different to my less informed view very low and the chance that the experience will make me angry and sad very high, is it then worth doing?
I want to be able to engage with opinions that I disagree with but for some opinions I find this very challenging. I am often wondering about how hard I should push myself here. Any thoughts on that?
I don't have a ChatGPT subscription. If I stop using their free tier, I think this has two effects: It benefits them because they can spend less resources on inference and it hurts them because investors lose trust in them. Do you think the expected net effect is positive or negative? Should I stop using their free tier if I want to protest against OpenAI?
Also, I noticed that I can check multiple boxes on https://quitgpt.org/ and only one of them is "I cancelled my subscription". If I understand the footnote below the number displayed at the top (currently 1,200,000) correctly, it counts everyone who checked any of the boxes. I would be more curious about how high the number of only the people is, who canceled their subscription or commit to not paying for ChatGPT in the future.
The standard comment sorting ignores what I have set in the settings as standard comment sorting and instead always sorts by New & upvoted by default.
I might have found a regression with the new post page when reading sequences: The "previous post" and "next post" buttons on the bottom are missing. For example in this post. Only the left and right arrows on the top and the possibility to go to the sequence overview between the two arrows are still there. I have also spotted this post, where the part at the top is also missing. It should be part of the same sequence.
This is a little outdated. The progress in AI is so fast that texts about AI capabilities become obsolete quite quickly.
AIs are now able to pass the Turing Test, win the International Math Olympiad, solve math problems that haven't been solved by anyone before, and produce a significant percentage of all new code.
Also, the number of people working on AI capabilities and AI safety have both increased significantly. I don't know if the ratio between them is still similar, but my guess would be that it is.
Thanks for this post. I especially found your point on treating the post-FTX challenge as a trust problem valuable. I always had a slightly bad feeling about this trend of trying to minimize reputational risks and distancing oneself from stuff that others might find controversial. After reading your post I have now more clarity regarding that.
Focussing on outward appearance comes with the risk of losing intellectual honesty and might not even be very effective at improving outward appearance as it is only treating the symptoms and might even seem deceptive or as if we had something to hide. I am not saying that one shouldn't care about outward appearance at all, but maybe it shouldn't be the main focus of dealing with the post-FTX challenge but rather thinking about it as a trust problem should be. The main priority should be to ensure that our community is worthy of our trust and that it creates value. I am aware that this is already happening to some degree, and I am not sure if it is possible to judge how much of it is by looking how much is publicly talked about it is, but if it is then outward appearance of EA is currently too much of a focus and making sure that EA is trustworthy, honest and true to its principles maybe deserves a little more attention.
Having a community worthy of our trust and one that is true to its original principles gets harder the larger the community gets, and it also gets harder when the community grows faster. I get the point that growing slower comes at a cost and more good can be done with a larger community but when slower growth means a healthier community that exists in a meaningful way for longer, then maybe that is a price we should be willing to pay.
Duncan Sabien's "Make more Grayspaces" is also interesting in this context. It describes a possible solution for how a growing community can stay true to its values. I am not sure if this solution would work for EA, but I found the description of the problem quite compelling.
I did, but I must confess that it wasn't for me. For some reason I found it really annoying to hear the two AI voices so often talk about their reactions or how they "felt" about a particular part of the book. I already find it annoying when humans do that instead of talking about the actual topic but when AIs fake that to appear more human like I find it quite irritating.
Sorry, I meant to write "ideas are more uncomfortable when they are new". Not because.
Yes, that is what I meant.
Ok, so your point is that ideas are more uncomfortable when they are new and when I already heard them, I'll find them less uncomfortable?
Not yet. To give the podcast the best chance, I'll start with a book summary that I think I'll enjoy. Once I've finished it, I'll decide how to proceed.
I am not sure what your point is regarding encountering those opinions in the wild. After listening to the book summary, it doesn't get easier to control how long I immerse myself in this in a conversation. But I don't think that is what you mean.
On your website you say that your goal with this project is to enable people to have their own informed opinion about a book so that they don't just repeat headlines. This is obviously a good thing, but I was skimming through the list of books and I must admit that for some my first reaction was "ok, you'll need to duraniumman that author's position to make me update my uninformed opinion on that book!" Of course, it is important to have a strong foundation on which to base a strong disagreement with a view. I am just not sure if I can force myself through it.
What is your view on this: If I rate the chance that my more informed view will be different to my less informed view very low and the chance that the experience will make me angry and sad very high, is it then worth doing?
I want to be able to engage with opinions that I disagree with but for some opinions I find this very challenging. I am often wondering about how hard I should push myself here. Any thoughts on that?
I don't have a ChatGPT subscription. If I stop using their free tier, I think this has two effects: It benefits them because they can spend less resources on inference and it hurts them because investors lose trust in them. Do you think the expected net effect is positive or negative? Should I stop using their free tier if I want to protest against OpenAI?
Also, I noticed that I can check multiple boxes on https://quitgpt.org/ and only one of them is "I cancelled my subscription". If I understand the footnote below the number displayed at the top (currently 1,200,000) correctly, it counts everyone who checked any of the boxes. I would be more curious about how high the number of only the people is, who canceled their subscription or commit to not paying for ChatGPT in the future.