Posts
Comments
The study doesn't measure people being charmed in the sense that they are perceived to be charming by other humans.
If you get someone to interact in style X that's normally a signal for Y it stops being a good signal for Y. That's the basic Goodhard's law principle.
That's not how the OP proposes this information to be used. All three paragraphs at the end that start bolded are recommending other ways to use the information.
From a male perspective a core way of expressing interest is to ask for a way to follow up and follow up.
From a female perspective signaling interest is more important, but engaging in actions believed by guys to be a show of interest is something different then engaging in actions from which a well-trained machine learning model can infer interest.
This sounds to me like it goodharts on the wrong thing. When on a date your core concern isn't to signal to the other person that you are flirting but that you are a desireable mate.
You are judging curi and FI (Fallible Ideas) via your standards (LW standards), not FI's standards. I think this is problematic.
The above post explicitely says that the ban isn't a personal judgement of curi. It's rather a question of whether it's good or not to have curi around on LessWrong and that's where LW standards matter.
Unpopularity is no reason for a ban
That seems like a sentiment indicative of ignoring the reason for which he was banned. It was a utilitarian argument. The fact that someone gets downvoted is Bayesian evidence that it's not valuable for people to interact with him on LessWrong.
How is this different to pre-crime?
If you imprision someone who murdered in the past because you are afarid they murder again, that's not pre-crime in most common senses of the word.
Additionally even if it would be, LW is not a place with virtue ethics standards but one with utilitarian standards. Taking action to prevent things that are likely to negatively effect LW from happening in the future is perfectly fine with the idea of good gardening.
If you stand in your garden you don't ask "what crimes did the plants commit and how should they be punished?" but you focus on the future.
As I envision the course, however, it would be an element such as a warm up or cash out, not the core curriculum.
Yes, however it's worth noting that you can play Zendo in different ways. The time I played it was with arrangement of Lego stones.
You could also play it with other domains like words or sentences. I'm uncertain about how much time a child can effectively learn something from Zendo.
At the same time it won't fill the full curriculum.
Focusing mostly because it seems like it would require one-on-one instruction, at least initially.
You didn't really speak about how many children you actually want to teach at one time.
When it comes to teaching rationality to adults there's the CFAR handbook. If you are not aware of it, it might be worth looking through it and thinking about what you can teach children.
When teaching it to a group of children I would likely not do it via an app but with physical items and drawings. It makes sense to switch the kind of items with which you play it between sessions to increase the generalizability of the learning.
Random thoughts:
Focusing, Internal Double Cruz and Belief reporting are powerful tools for aligning system I and system II. They aren't very complex and have relatively few moving parts. It might be possible that you can teach those to children but I could also imagine that it's hard to teach them. Dealing with groundlevel thinking is intimite and it needs buy-in from the child.
Zendo seems to be a game that can be fun for children and it teaches valuable lessons about building scientific hypothesis and testing them.
Credence calibration doesn't seem very complex but might be too much for a 7/8 year old. I'm unsure at what age the necessary understading of numbers between 1 and 100 exists.
I could imagine a setting where you talk with children about what they are curious about and what might be tested by experiment. Then the children gather data collaboratively and you do the math for them in Numpy.
If you have a humidity, temperatur and a CO2 sensor you can for example have experiments about how much opening the window affect the air in the room. Does open the window fully for 5 minutes do more then have it half open for 10 minutes?
Opening the windows to get better air is a task that's relevant to daily life and it's not a question where you find the answer in normal textbooks.
We want to teach children to accept the norms of our society and the narrative we tell about it. A lot of what we teach is essential pro-system propaganda.
Teaching moral uncertainty doesn't help with that and it also doesn't help with getting students to score better on standardized tests which was the main goal of educational reforms of the last decades.
most people
That's overstating it. Most people don't live in the US and are not bound by it.
Compared to (good) software developers, lawyers do not seem to be very good at this; they tend to throw patches on top of patches, creating more corner cases rather than fewer.
There are plenty of different lawyers and different one's have different interests.
YCombinator for example has a bunch of relatively short contract documents on offer. YCombinator has an interest in startup short contracts in a way that most law forms who pay for their legal services don't.
Huh, interesting. That's surprising to me; I expected contracts to have a sufficiently long history that there wouldn't be any recent major innovations.
The key change doesn't seem to be so much about innovation but about standardization between different jurisdictions.
Standardization between multiple jurisdiction that each like their own standards is hard as shown by the US still using a lot of imperial units.
Thanks.
Nobody has called me an anti-vaxxer. Generally, I'm more concerned with giving anti-vaxxers who don't care for science more ammunition then I'm concerned with being called that term.
The danger of people dying because I wasn't careful with qualifying criticism is much more serious then getting called a name.
That shouldn't keep us from doing our best to find out the side effects in the time we have
How serious are you? What sounds to me like babble.
Calling for doing our best basically means doing a lot of human challenge trials for which people like me and Zvi called for months ago. The current consensus is that most people don't want to do our best to search for side effects in the time that we have but rather want to follow traditional research processes.
To me this seems like a partisan label which just makes it harder to voice legitimate concerns about vaccines.
I don't think that's the case and I do happen to be a person who raised legitimate concerns about vaccines in the past.
do we know the side effects?
That's not a good question. We never know all the side-effects. We make decisions in uncertainty and have to think about the expected value of our decisions given the uncertainty that we have.
That said, AstraZeneca abandoning a vaccine for one patient with an adverse reaction seems absurd.
Clinical trials are a heavily regulated domain which rules that are not made for fast development of new drugs.
Even when death plummed, we still have some evidence that suggests that 3/4 of the infected people will have permanent damage.
I would prefer headlines to spell out the terms they use to make it clearer to a reader that scans the headlines what a post is about.
https://www.condairgroup.com/humidity-health-wellbeing/scientific-studies/criteria-for-human-exposure-to-humidity-in-occupied-buildings shows how the effect on humidity on various risk factors and based on it I would suggest 50-60% is ideal for fighting viruses.
That's a bit like saying: "What are all those AI safety people talking about? Can you please give me three specific examples of how they propose safety mechanisms should work?"
I haven't seen easy answers or a good link for them. At the same time, the project is one of the answers for the question in the OP.
My advice to employees would be to be very hesitant about coming to management with a complaint that you can’t/don’t want to do a core part of your job.
Ugh fields usually come up when people are faced with new challenges that they are not used to dealing with. Plenty of times that's not a core part of the job.
While I can see that there are plenty of people who don't like report writing and would rather prefer to do other things, it's not my model that most people have ugh-fields around them (that it's very painful to think about doing the task).
Is your model of people different or are you overgeneralizing the term?
I'm not sure what the best point of entry is. Youtube videos like https://www.youtube.com/watch?v=HL5bcgpprxY do give some explanation.
It sounds like they do an advanced form of Dual-N-Back. Gwern's analysis suggests that the reported effects of Dual-N-Back don't reproduce and are artifacts.
There's the Game B discourse around creating social norms that defeat moloch.
I think there's an idea that a paper with a p=0.05 finding should replicate 95% of the time. If it doesn't then the p-value was wrong.
Maybe we need a yearly award for the scientist who cites the most redacted papers?
Room humidity matters for COVID-19 transmission. If you are going to spent a lot of time in the same rooms as other people in the next months, invest into proper humidity to reduce your risk of getting ill: https://aaqr.org/articles/aaqr-20-06-covid-0302?fbclid=IwAR3zFZ-UqSjBlc2DJUjHI5yUKTujIW5WyDlwgogmfAAIJtEAxoCas-LkdWc
There's the other externality of becoming dependent on the bridge existing. Right now supply chains that were working well a year ago collapsed as COVID-19 lead to borders being closed.
Schools around the world seem to start using automated grading for tests. If that technology exists, it would be interesting to have a forum that enforces posts to have a minimum score on those grading forms.
I integrated information about phages from your link into my article.
10% is a big deal but there are many different ways to persue better health outcomes and if we overrate particular approaches and underfund others that can be a problem.
At the moment we have 35 vaccine candiates in clinical evaluation for COVID-19 while on the other hand we have trouble spending 20 million to have decent studies on whether the top 8 most promissing antivirals helps with COVID-19.
In theory you can overdose on Vitamin D, but it’s something that can only be done on purpose with extreme supplementation.
This is only true if you take K2 along with your Vitamin D. I know a fellow rationalist who thinks they overdosed on 5000 IU Vitamin D when they took it without K2.
See also: https://www.lesswrong.com/s/5CNs9wmHWFQTNjFKo/p/c5aycbSsSc38XWPEc
Canonical works on getting Flutter to run well on Linux: https://medium.com/flutter/announcing-flutter-linux-alpha-with-canonical-19eb824590a9
The rending engine should be Skia in a GTK+ shell. There's no intermediate product of HTML when you run Flutter on Linux.
Windows and Linux support made this year into the main Flutter repo, so there aren't yet many examples but there's a lot of work going into making Flutter great on the desktop.
When starting a new project like the one you are proposing I think there's a good chance that Flutter is the best platform even when the desktop support is still in it's early stages.
https://www.reddit.com/r/FlutterDev/comments/iqnwo5/what_are_good_flutter_desktop_apps_that_currently/ gives a list of existing desktop apps with Flutter.
Does anybody know the easiest way to get Google Home to record an idea?
Level 2: reusable rockets, space planes; ~$100 per kg to LEO
The economics that Elon predicts are more ~$10 per kg to LEO then ~$100.
There aren't any great alternatives to the web for multiplatform UI development, afaik, and the site will need to be able to render a lot of things for the web anyway, for the sake of making it widely accessible.
Flutter more and more becomes that platform.
The problem with this scenario is the part where the AI “solves the protein folding problem”. That the problem is NP-hard means that it will be difficult, no matter how intelligent the AI is.
There's no good reason to assume protein folding to be NP-hard. DeepMind seems to make good progress on it.
That makes sense as an upside, but it doesn't explain why they'd use Google Docs, because you could just have the candidate program in a normal text editor and then copy-paste it into Google Docs afterwards.
It's plausible that they not only care about the final code but also the process of writing the code. Simple copy pasting might not give them that data. On the other hand their custom build editor for interviewing.google.com might provide it.
As far as I understand it, to have your awareness buffer not filled by your mind running you need to fill it with something else.
Meditation is a way to learn to fill your attention with what you perceive. Various physical activity also fills awareness buffer.
Is this not the kind of topic where there's a consensus. If you want further arguments I wrote a long post on Vitamin D for LW: https://www.lesswrong.com/s/5CNs9wmHWFQTNjFKo/p/c5aycbSsSc38XWPEc
This includes all sort of material processes (though, of course, gravity makes a great many things easier), and the lack of environmental concerns.
If you want something like gravity you can easily get it by rotating your space station. Whether you want less or more then there is on earth space stations can easily get it.
You can create your crystals in zero G and then fly them to a earth-like pseudo gravity to give them to mice for animal testing.
When it comes to your wider argument I don't see an articulated reason for why we should go to the moon. You didn't lay out anything that space stations can't provide.
You need more then one launch to get to the moon with Starship as it's designed. It needs refulling in between.
Even when it costs 15k to low-earth orbit you might pay 100k to the moon.
I have not tested since I wrote the post and I made the arguments I had thought out in the post.
The other is frequentism, which could be simplified (strawmanned?) as "the situation must happen many times, and then 'probability' is the frequency of this specific outcome given the situation".
That definition has the advantage of defining probability as something that's objective while the Bayesian definition depends on the prior beliefs of a particular person and is subjective.
The Finnish user in early march is likely Petri Hollmén. I wrote a bit about him back in Using the Quantified Self paradigma for COVID-19 . Petri mainly got convinced about the decision to get tested mainly based on a nightly temperature rise from Wednesday to Thursday when he was infected on the weekend.
Unfortunately, Petri wasn't willing to share resting heart rate and heart rate variance (HRV) so I don't know. I and others asked him on his Facebook post where he described the episode but he didn't reply.
But what CFAR found was that humans are actually really really good at Bayesian reasoning, when other psychological factors are not getting in the way.
What's the source for that claim? Is that a public position of CFAR?
I actually don't really know how to do the math on this one. If we start out with a population that all has the normal form of gene B and we the mutant form conveys no net reproductive fitness benefit or downside and the likelihood of each mutating into the other is equal, then I suppose we would expect the frequency of each variant to approach 50% given enough time.
The relevant math would be Gambler's ruin. If two forms have equal and independent chances of reproduction over a longer time-frame either of the forms will be wiped out.
But when it comes to the core of evolutionary theory, there isn't even a good reason to try to do original research. There are plenty of professors in evolutionary theory that spend a lot of time investigating the issue, so unless you have good reason why the mainstream professors in evolutionary theory are wrong, it makes sense to default to the mainstream academic beliefs.
The unit for your 30k seems to be BPEs (Byte pair encodings).
I found on https://www.gwern.net/GPT-3#dialogue:
The first limit is that it remains hobbled by the limited context window. GPT-3 has no form of memory or recurrence, so it cannot see anything outside its limited 2048 BPEs
If GPT-2 could have a context window of 30k BPEs with 300GB ram, could GPT-3 also have such a context window length? So it could be made 15 times as big as it's currently?
I don't think expecting readers to read a bunch of long post before engaging with political posts on LessWrong is a reasonable demand. Furthermore you didn't use the terms as defind in the posts you referenced.
I think it's reasonable to say "I assume X to be true because argument from Y source" but that's not how your post goes.