Posts
Comments
Hi! I wrote two extensions you suggested:
- "Emotion highlighter" detecting and highlighting paragraphs with 6 basic emotions
It's very basic API call right now, I'll think about improving it once I see if anyone uses it at all and what improvements they want (more emotions / more precise highlighting / better classification?).
- "Simple English translator" converting all text on a webpage into plain English.
They use your OpenAI API key to analyze all text on a webpage once you click the extension, and OpenAI charges $5.00 / 1M input tokens as of the time of writing this comment.
>> High decouplers will notice that, holding preferences constant, offering people an additional choice cannot make them worse off. People will only take the choice if its better than any of their current options.
This is demonstrably untrue in cases of suicide. 70% of people who survive a suicide attempt do not attempt it again, so their decision to try is probably a bout of temporary madness / irrationality, and not an expression of stable well-considered preference for death over life.
Long Live the Queen takes about 4 hours. It would take some luck to beat it on the first try, but generally you win by using common sense and training useful skills.
Have you read https://www.lesswrong.com/posts/NEeW7eSXThPz7o4Ne/thou-art-physics ?
What were the most impressive results that were supposedly accomplished? Maybe we should run some experiments to see if they can be reproduced?
Interesting post and self-improvement advice!
I want to add to that when you're learning a behavior that is new to you, you want to start out in a low-risk setting. You are probably going to make mistakes. For example, when you first try to act high-status and assertive, you may overdo it and come off as aggressive and rude. So you want to start practicing in a safe setting where mistakes won't cost you a job, e.g. anonymously on the internet or with friends. Or if you do write to your boss, have a friend look over the letter. Then move on to more high stakes situations.
Another caveat is that if you adopted a dysfunctional behavior, it was probably for a reason. There was probably someone in your environment around whom the dysfunctional behavior made complete sense and was the only way to get along with them. They may still be around you. Expect them to blow up. But don't be discouraged, remember most people are not like them.
I appreciate you writing this post! I was curious about the evidence for lab leak, but was too lazy to investigate on my won.
You point out COVID-19 is the only sarbecovirus with furin cleavage site. But couldn't it have evolved by switching host from some other species? According to Nature, "viruses more often evolve by jumping from one host species to another than by remaining within a particular species."
So the general prevalence of furin cleavage sites seems relevant too. Did anyone look into what that is?
Didn't they demonstrate that transformers could be mesaoptimizers? (I never properly understood the paper, so it's a genuine question.) Uncovering Mesaoptimization Algorithms in Transformers
Not to mention that the default result of rebellion is failure. (Figure from https://www.journalofdemocracy.org/articles/the-future-of-nonviolent-resistance-2/)
OP doesn't claim that dictators are unchecked in their power, he jokingly claims that dictators and monarchs inevitably end up overthrown. Which is, of course, false: there were ~55 authoritarian leaders in the world in 2015, and 11 of them were 69 years old or older, on their way to die of old age. Dictator's handbook has quite a few examples of dictators ruling until their natural death, too.
https://foreignpolicy.com/2015/09/10/when-dictators-die/
I agree that it's his main point; however, he's also making an observation that most terrorists are incompetent, impulsive, have poor preparation and planning, and choose difficult forms of attacks when better options are available. The post has several anecdotes illustrating that.
He believes the incompetence is caused by terrorist acting on social incentives instead of optimizing for their stated goals. However, what if some terrorist group has one earnest terrorist, or what if the chatbot provides the social encouragement need to spur a terrorist to action while simultaneously suggesting a more effective stategy? There also lone wolf terrorist who, while more practical, are limited to their own ideas, so probably less competent than a whole team of researchers.
Ok, so LLMs don't give an advantage in bioterrorism planning to a team of RAND researchers. Does it mean they don't give an advantage to actual terrorists, who are notoriously incompetent? https://gwern.net/terrorism-is-not-effective#sn17
I think you're zeroing in on the hypothesis that your list had a problem too early. There are many possible reasons to fail college, like having a mental health issue or not being very motivated to succeed in the first place. Do you know that he actually read your list?
In my experience the biggest predictor of teaching success is whether the person asked to be taught)
If you are leaving relatives or friends behind, consider developing some kind of code language, because people in Russia might be afraid to tell you their real opinions over the phone in plain speech.
I believe there is nothing wrong or irrational about taking collective action or calling to it. On the contrary, a culture that prohibited collective action has failed at instrumental rationality and is about to be conquered by a culture that didn't. So I am strongly opposed to your first suggested rule.
Yes, I believe we shouldn't get involved in politics if it endangers alignment research efforts or otherwise hurts the community for little gain. But we should take collective actions which carry negligible risks and huge expected benefits. Rationality is about winning. Being divided makes us weak and less likely to win. Let's not do this.
Will MIRI want to hire programmers once the pandemic is over? What kind of programmers? What other kinds of people do you seek to hire?
I agree that this statement could be understood this way, and I don't find your interpretation objectionable. It also could be understood to mean that Russian POWs say what they say to stop torture, there is no disclaimer against this interpretation. I should probably have interpreted everything in the most charitable way possible, if it was one thing. I am pushing back because several things made me feel paranoid.
Advising Ukrainans to flee while banning commenters from giving any advice to Western powers or discussing morality and justice seemed not neutral. (Russian people remaining nonviolent is not in Ukrainians' interest either.) Comparing Putin's situation to a trolley problem frames him as someone selflessly trying to do what's best for others. I am not saying that these posts are pretending to objectivity while secretly being kremlin propaganda, it just looks to me like lsusr is trying to be neutral and falling short.
Thanks for the report.
If I was captured by America's enemies I'd happily shout "Death to America!" on camera in exchange for humane treatment.
You're implying that Russian POWs are treated inhumanely unless they say what the captors want. That assertion needs proof.
There is also this isolating effect when the media tells a lie, e.g. denies some true fact X, and most people don't buy it, but they decide "this is what we're all conspiring to tell our enemies". And while you're busy trying to "convince" your opponent that X actually happened, you never get to discuss whether X is the right thing to do.
Wait, I thought EA already had 46$ billion they didn't know where to spend, so I should prioritize direct work over earning to give? https://80000hours.org/2021/07/effective-altruism-growing/
It seems like the source of your disagreement is that you do not believe turkeys actually suffer (as you write "suffer" in scare quotes), while the OP clearly believes they do. I think this question needs to be settled first before we decide which emotional reactions are reasonable. (I myself have no idea what the answer is.)
Radical actions. The word "radical" means someone trying to find and eliminate root causes of social problems, rather than just their symptoms. Many people pursue radical goals through peaceful means (spreading ideas, starting a commune, attending a peaceful protest or boycotting would be examples), yet "radical act" is commonly used as a synonym to "violent act".
Extremism. Means having views far outside the mainstream attitude of society. But also carries a strong negative connotation, in some countries is prohibited by law and mentioned alongside "terrorism" like they're synonyms, and redefined by Wikipedia as "those policies that violate or erode international human rights norms" (but what if one's society is opposed to human rights?!) Someone disagreeing with society is not necessarily bad or violent, so this is a bad concept.
"Outside of politics". Any choice one makes affects the balance of power somehow, so one cannot truly be outside. In practice the phrase often means that supporting the status quo is allowed, but speaking against it is banned.
What the heck did I do wrong, why are you downvoting me, guys?
Are you saying that MIRI enforces altruism in their employees? If so, how do they do that, exactly?
Scott Aaronson, for example, blogs about "blank faced" non-self-explaining authoritarian bureaucrats being a constant problem in academia. Venkatesh Rao writes about the corporate world, and the picture presented is one of a simulation constantly maintained thorough improv.
Well, I once met a person in academia who was convinced she'd be utterly bored anywhere outside academia.
If you want an unbiased perspective on what life is like outside the rationality community, you should talk to people not associated with the rationality community. (Yes, Venkatesh Rao doesn't blog here as far as I can tell, but he is repeatedly mentioned on LW, so counts as "associated" for the purpose of this exercise.)
How do you know it's useful though? Did you apply the advice? Did it help?
>> Sometimes I try to tell the people what I can see, and that doesn't always go well. I'm not sure why.
Can you describe a concrete example? Without looking at a few examples, it is hard to tell if a "context-free integrity" fallacy is to blame, or you are just making bad arguments, or something.
One benefit is that you can do calisthenics everywhere (even in prison), no need for special equipment.
What's to stop the prosecutor from lying about their Briers score?
We have exercised our innovative technique of meta-honesty to successfully dupe some computer programmers into thinking we want their participation.
I actually thought Ty was a real person. :)
And Y/2 pain, probably? (Or the conclusion doesn't follow.)
How do you measure intelligence though? Obviously you don't mean IQ, since IQ test scores are deliberately calibrated to be normally distributed.
Generally, I feel like all these bold claims need some supporting evidence. E.g.
Increased information makes smart people smarter and stupid people stupider.
Citation needed?
I totally agree that it's useful to hang out with a diverse set of people.
It also helps to treat people's opinion of you as an instrumental goal. Every time I'm worried what someone thinks of me, I ask myself if this person's opinion is important, and why - can they hurt me or help me in any way? Sometimes the answer is yes, e.g. I want to impress employers, or I need voters to like me if I'm doing politics. Often, though, the answer is that the person is not going to affect my life in any way, and so their opinion doesn't matter. People's opinions may also matter as an estimate of my own virtue, but if their opinion is based on a misunderstanding, or they're confused about what's virtue and what's vice, then their opinion can be discarded again.
Is starting capital really a bottleneck for entrepreneurs? Don't you just get money from investors?
Elon Musk and Bill Gates only needed a laptop to start their business. Or, from Warren Buffet's biography: "In 1945, as a high school sophomore, Buffett and a friend spent $25 to purchase a used pinball machine, which they placed in the local barber shop. Within months, they owned several machines in three different barber shops across Omaha. They sold the business later in the year for $1,200 to a war veteran. ... In high school, he invested in a business owned by his father and bought a 40-acre farm worked by a tenant farmer. He bought the land when he was 14 years old with $1,200 of his savings. By the time he finished college, Buffett had accumulated $9,800 in savings".
Were long-standing research problems actually ever presented at IMO? AFAIK, problems featured there already have solutions.
I feel like I'm missing context. Why did this community come to care about blackmail laws in the first place?
I'm so confused. How did Luna survive the Killing Curse?
Great observation! I was struggling with the same issue when I moved from studying math to graduate ML research. Depth-first search is the right approach to reading a math textbook. Say, you started to learn homology theory and realized you don't know what's an Abelian group. You should stop and go read about Abelian groups, or you won't understand what comes next.
However, the same approach was getting me in trouble when trying to understand state-of-the-art in voice processing. I would start reading an article in the morning, and by the evening I'd finish no articles and find myself stuck in the middle of a textbook reading about some 50 years outdated method which wasn't relevant to modern research. So, I worked out that the effective approach to this task is breadth-first search: read the article from beginning to end, write down all unknown terms, go on to look up the most important one.
I think another reason why people might default to depth-first search is that BFS requires to store a list of unexplored nodes in memory and people don't have a lot working memory. So a note-taking system like Zettelkasten really helps with applying this approach more broadly in one's life.
Do you know anyone who wants such a forum?
Cheers to simon, ericf and myself, for offering an optimal solution! And cheers to abstractapplic for organizing the challenge.
The leaderboard (if you're not here, I couldn't figure out what your final decision was, or you added more than 10 points):
simon, ericf 0.9375
[('CHA', 8), ('CON', 15), ('DEX', 13), ('INT', 13), ('STR', 8), ('WIS', 15)]
seed 0.9375
[('CHA', 8), ('CON', 14), ('DEX', 13), ('INT', 13), ('STR', 8), ('WIS', 16)]
Samuel Clamons 0.8095
[('CHA', 8), ('CON', 17), ('DEX', 13), ('INT', 13), ('STR', 7), ('WIS', 14)]
Asgard 0.7857
[('CHA', 9), ('CON', 16), ('DEX', 14), ('INT', 13), ('STR', 8), ('WIS', 12)]
Measure 0.7308
[('CHA', 8), ('CON', 14), ('DEX', 13), ('INT', 13), ('STR', 6), ('WIS', 18)]
kiwiakos 0.6774
[('CHA', 7), ('CON', 15), ('DEX', 13), ('INT', 13), ('STR', 6), ('WIS', 18)]
Alexey 0.6500
[('CHA', 11), ('CON', 14), ('DEX', 13), ('INT', 13), ('STR', 6), ('WIS', 15)]
newcom 0.6471
[('CHA', 11), ('CON', 16), ('DEX', 13), ('INT', 13), ('STR', 7), ('WIS', 12)]
AABoyles, Pongo, GuySrinivasan 0.6389
[('CHA', 6), ('CON', 14), ('DEX', 13), ('INT', 13), ('STR', 6), ('WIS', 20)]
Yongee 0.6364
[('CHA', 5), ('CON', 14), ('DEX', 13), ('INT', 20), ('STR', 8), ('WIS', 12)]
Deccludor 0.6098 [('CHA', 5), ('CON', 20), ('DEX', 13), ('INT', 13), ('STR', 6), ('WIS', 15)]
Randomini 0.4688 [('CHA', 4), ('CON', 14), ('DEX', 13), ('INT', 13), ('STR', 16), ('WIS', 12)]
From plotting the data, I saw that:
- looked like the stats were independtly randomly generated and cut off at sum >= 60.
- dexterity was useless
- looked like there was a big advantage to having stats >= 8. For strengh, strength=8 was almost as good as strength=20
I fit a regularized logistic regression and a neural net, but couldn't get validation accuracy greater than 70%, which was only a little better than the 65% baseline of random guessing. I realized that the data is not very informative and I don't know how results are calculated, so I better stick with a conservative model like Nearest Neighbors classifier, and try a few different models. I fit a KNN classifier, gradient boosting on decision trees, and regularized logistic regression (all with validation accuracy 70%), and chose a point which gave near the top scores for all three classifiers. (It had all stats >=8, too.)
I wonder if it has something to do with age-related hearing loss. I remember when I was a teen, rock music sounded like antimelodic screeching to me too, but I listened because my friends liked it, and I liked the lyrics. Now the same songs sound like legit music to me. Maybe it's because our hearing range shrinks with age, so the high-pitched sounds become quieter and the song doesn't sound like screeching anymore.
Is it a secret / part of the puzzle, where this data came from?
Do adventurers gain additional status points during their Great Quest, and if yes, are these stats measured at the beginning or at the end of their quest?
Is this data from some real Dungeons and Dragons game results?
This was very funny and the best HPMOR continuation I've read so far.
Oh, sorry, I've only heard the word used in that context before, I thought that's what it meant. Turns out it has a broader meaning.
Look, I'm neurotypical and I don't find anything Eliezer writes offensive, will you please stop ostracizing us.