Unnamed

Karma: 6,951

Unnamed 2 Aug 2022 23:19 UTC
4 points
0 ∶ 0
in reply to: Valentine’s comment on: Flash Classes: Polaris, Seeking PCK, and Five-Second Versions
Seeking PCK was a full (hour or longer) class at every mainline workshop since October 2016 (sometimes called “Seeking Sensibility” or “Seeking Sense”). After you left it was always a full hour+ class, almost always taught by Luke, and often on opening night.
The concept of PCK became part of the workshop content in April 2014 as a flash class (as a lead-in to the tutoring wheel, which was also introduced at that workshop). In October 2016 we added the full class, and then a couple workshops later we removed the flash class from the workshop. Something very close to this chapter made it into the first draft of the CFAR handbook in May 2016, when PCK was still just a flash class, and I guess the chapter didn’t ever get expanded or moved.
After the class was transferred from Val to Luke, Luke was involved in teaching it until the last pre-covid workshop in January 2020. I’m pretty sure he kept the subtraction exercise (that exercise wasn’t removed from the handbook, it just never made it in). A couple other people also taught the class at some point (including Duncan in April 2017), I’d guess at workshops where Val or Luke was absent.
At the January 2020 workshop a new instructor was learning the class & taught some of it along with Luke. I suspect that’s why it was part of the day 1 rotation that workshop rather than being opening night (since it’s helpful for a new instructor to have repetition & smaller groups, and a new instructor’s version of a class hasn’t necessarily cohered enough to be ready to set the tone for the workshop on opening night).
(This history mostly based on records I looked up, supplemented by memory.)

Unnamed 12 Jul 2022 19:51 UTC
10 points
5 ∶ 0
on: Marriage, the Giving What We Can Pledge, and the damage caused by vague public commitments
I would click the “disagree” button if there was one, because many parts of this post are askew to how I understand marriage, divorce, commitment, etc.,
I think of a marriage as two people deciding to build a life together, and commitment as essentially about being “in” on that shared project. This post seems to be coming at it from a different angle, where explicitly specifying things in advance is much more fundamental. It centers honesty vs. dishonesty, ironclad promises, and public accountability in places where those don’t feel like the central concepts to me.
A few of the places where that disagreement came up most strongly:
- The paragraph about marriage beginning “The point of a public pledge is help structure our own incentives to fulfill our commitment.” That does not seem like the main point of having a public ceremony, which IMO is more about marking the occasion of the couple building a life together, in common knowledge so that their surrounding community will treat them as a unit.
- The paragraph about the mountain climbing promise. Agreeing to climb a mountain before a divorce is finalized doesn’t actually seem to help with the things this post is describing as important. This promise mainly just seems to be acknowledging that an escape clause (divorce) already exists, and adding in an extra step which perhaps is symbolically meaningful to the couple.
- The partial sentence “I want people to build within themselves the machinery to be able to make strict pledges that mean things”. As if a pledge needs to be strict in order to be meaningful. There are plenty of meaningful marriages which didn’t specify in advance criteria for a divorce or social consequences of a divorce.
The complaints in the GWWC Pledge section also don’t seem that similar to the arguments in the section on marriage. Standards like ‘Don’t say false things’ (like “till death do we part”) and ‘make it possible to opt out later on, and acknowledge this possibility up front’ which seemed like central parts of the section on marriage are already covered by the current version of the GWWC Pledge.

Unnamed 7 Jul 2022 20:41 UTC
4 points
0 ∶ 0
on: When Giving People Money Doesn’t Help
The predicted effect sizes (.16-.65 SDs) seem too large, compared to (e.g.) the size of the linear relationship between log(income) and well-being that many studies find. I would’ve expected a positive effect rather than zero or a negative effect, but probably on the low end of that range or below it, depending on the specific question.

Unnamed 5 Jul 2022 23:21 UTC
4 points
3 ∶ 1
on: ITT-passing and civility are good; “charity” is bad; steelmanning is niche
I think I basically agree with Rob about the importance of the thing he’s pointing to when he talks about the importance of “Trying to pass each other’s Ideological Turing Test”, but I don’t like using the concept of the ITT to point to this.
It’s a niche framing for a concept that is general & basic. “Understand[ing] the substance of someone’s view well enough to be able to correctly describe their beliefs and reasoning” is a concept that it should be possible to explain to a child, and if I was trying to explain it to a child I would not do that via the Turing Test and Bryan Caplan’s variant.
The Ideological Turing Test points to the wrong part of the process. Caplan says that the ability to pass ITTs is the ability “to state opposing views as clearly and persuasively as their proponents.” This means that the ITT is a way of testing whether a person has that ability to state opposing views well. But what we care about in practice is whether a person has that ability & is using it, not how they do on this particular way of testing whether they have that ability. This kind of imitation test usually doesn’t come into play.
There have been several ITT contests, where authors write short essays about either their own views or the views of an ideological opponent, and guessers try to tell which essays represented the authors own views. The vibe of these contests doesn’t match the thing that matters in conversation, and ‘pretend to be a proponent of X’ can be pretty different than understanding a particular person’s views. The contests involve pretending to be something you’re not, they involve capturing the style and not just the substance, and they involve portraying a member of the group rather than getting a single person’s views.
We need better terminology for disagreements, arguments, and similar conversations.

Unnamed 1 Jul 2022 18:50 UTC
4 points
1 ∶ 0
on: The Track Record of Futurists Seems … Fine
I recall hearing a claim that a lot of Kurzweil’s predictions for 2009 had come true by 2019, including many that hadn’t happened yet in 2009. If true, that supports the picture of Kurzweil as an insightful but overly aggressive futurist. But I don’t know how well that claim backed up by the data, or if there even has been a careful look at the data to try to evaluate that claim.

Unnamed 21 Jun 2022 22:35 UTC
9 points
on: One single meme that makes you Less Wrong in general
If someone, somewhere, were to have a vested interest in keeping consumer spending high in order to stave off a recession, they would at least try find a way to persuade millions of people that There Is No Recession.
I think the basic story behind that WSJ headline is that the financial press makes an overly big deal out of daily market fluctuations which aren’t that relevant to most investors (or most people). Not that they’re trying to trick people into thinking that the economy is doing better than it is.
To pit these two hypotheses against each other (or other hypotheses), you could look at the past couple weeks of financial headlines to see what the WSJ & other financial press said on days when the market went down.

Unnamed 20 Jun 2022 7:55 UTC
2 points
on: What’s the name of this fallacy/reasoning antipattern?
Rationalization

Unnamed 18 Jun 2022 1:26 UTC
3 points
on: Humans are very reliable agents
I wouldn’t call the low death rate from surgery humans being highly reliable. Surgery used to be much deadlier. Humans have spent many many years improving surgical methods (tools, procedures, training), including by using robotic assistance to replace human activity on subtasks where the robots do better. Surgery as practiced by current trained humans with their tools & methods is highly reliable, but this reliability isn’t something inherent to the humans as agents.

Unnamed 16 Jun 2022 1:04 UTC
43 points
on: Contra Hofstadter on GPT-3 Nonsense
GPT-3 reminds me of a student bullshitting their way through an exam on Dave & Doug’s version of these questions. This question doesn’t make any sense to me, but I guess the teacher expects me to have an answer, so I’ll see if I can make up something that resembles what they’re looking for.
There is a mind-boggling hollowness hidden just beneath the flashy surface of a clever student who is just trying to guess what their teacher is looking for.

Unnamed 13 Jun 2022 20:43 UTC
6 points
on: Why so little AI risk on rationalist-adjacent blogs?
Matthew Yglesias has written a couple things about AI risk & existential risk more broadly, and he has also talked a few times about why he doesn’t write more about AI, e.g.:
I don’t write takes about how we should all be more worried about an out-of-control AI situation, but that’s because I know several smart people who do write those takes, and unfortunately they do not have much in the way of smart, tractable policy ideas to actually address it.
This seems different than your 8 possibilities. It sounds like his main issue is that he doesn’t see the path that you think you see where “Rationalist-adjacent writers are a major path for LessWrong ideas to influence elite and mainstream opinion. This can lead to good policies, like avoiding a race with China and discouraging certain types of capabilities research.”

Unnamed 27 May 2022 3:52 UTC
4 points
in reply to: alyssavance’s comment on: Is AI Progress Impossible To Predict?
Doesn’t that mean that you are getting some predictiveness by looking at momentum? If progress on a task was totally unpredictable, with no signal and all noise, then your way of carving up the data would produce negative correlations. Instead you’re mostly finding correlations near zero, or slightly positive, which means that there is just about enough signal to counteract that noise.
The signal to noise ratio is going to depend on a lot of contingent factors. There will be more noise if there are fewer questions on a task. There will be less signal from one model version to the next if there is a smaller increase in model size, or if the task is one where improvement happens very gradually as models scale up (though in those cases you could find a clearer signal by looking across several model versions, rather than just two consecutive jumps).

Unnamed 29 Apr 2022 6:44 UTC
6 points
on: Virtue signaling is sometimes the best or the only metric we have
I understand “virtue signalling” as a novel term that is only loosely related to the concepts of “virtue” or “signalling”.
It’s a little annoying to have to mentally translate it into “that thing that people mean when they say ‘virtue signalling’” (or sometimes a little amusing, when there’s an interesting contrast with the literal meaning that a person’s actions have signalled their virtues).

Unnamed 11 Apr 2022 7:02 UTC
8 points
on: Playing with DALL·E 2
- this photo was taken with perfect timing as the water balloon was breaking
- note the symmetry in this ancient Roman mosaic from the National Museum of Roman Art in Merida
- hot air balloons over the rolling hills of Tuscany
- literal puppy love, awwwww
- a sculpture at the Museum of Modern Art based on a child’s drawing of a monster
- traffic on the busy streets of Sydney

Unnamed 6 Apr 2022 23:00 UTC
2 points
on: Down By 30
Down by 30 I probably put in backups to make sure my best players don’t get injured, and play to run out the clock. Winning would require a record-breaking comeback, and even if we go all-out to win the chances of pulling it off are tiny, maybe 1 in a million.
Though I guess if it’s the playoffs then I keep playing for the win. Regular season it would be worth playing for the win if we’re down by 20 instead of 30.
Generally teams do adjust their tactics in the right direction in these sorts of situations, but not by enough on average. NFL teams play faster when trailing by multiple scores, but they usually don’t shift to the full-fledged hurry-up offense that they use when time is running low at the end of the half. Teams go for it more on 4th down, but there are still plenty of cowardly punts. Quarterbacks throw more interceptions due to trying to make more risky passes down the field into tight coverage—I guess I’m less sure about how to tell whether they do too little of that. We don’t see many trick plays in these situations—something like a flea flicker might not be that effective when the defense is playing it safe, but I expect that using more laterals (like a hook & ladder play) would be worth the risk (though I’m not sure about the cost-benefit of putting those on tape vs. saving them for higher leverage situations).

Unnamed 1 Apr 2022 8:59 UTC
−2 points
on: Replacing Karma with Good Heart Tokens (Worth $1!)
test post

Unnamed 31 Mar 2022 4:37 UTC
7 points
on: Confidence Levels in Forecasts and Psychological Surveys
Response bias or “response style” is the keyword for phenomena like these. “Extreme responding” or “extreme response style” are terms for gravitating away from the neutral option.
I don’t know if there’s research on whether this is correlated with overconfidence.

Unnamed 12 Mar 2022 23:19 UTC
7 points
on: Compute Trends — Comparison to OpenAI’s AI and Compute
It looks like the AlphaGo models play a huge role in the “trend” of large-scale models.
In your spreadsheet, AlphaGo Master (January 2017) was much larger than anything that came before it (22x the training compute). AlphaGo Zero doubled that later the same year (October 2017) and remained the largest model by training compute for almost 4 years, until August 2021. By late 2021 the trend had caught up and now 5 models have exceeded AlphaGo Zero, all language models, led by Megatron-Turing NLG 530B (October 2021) which has 4x the training compute.
The trend looked steeper when AlphaGo Master & Zero were near the end of the trend line, when OpenAI was analyzing it in 2018. The trend for the “large-scale era” looks shallower now that those models are in the first half of the era. To me it looks like those models are off-trend, even compared to other “large scale” models. DeepMind was much more willing to throw compute at the AlphaGo models than anyone has been with any other model.

Unnamed 4 Feb 2022 0:21 UTC
8 points
on: On Bounded Distrust
The masks story fits the template of bounded distrust. You sum it up here as:
Let me tell you a story, in three acts.
1. All masks don’t work unless you’re a health professional.
2. All masks work.
3. Cloth masks don’t work.
At each stage of this story, scientists got on television to tout the current line. At each stage of this story, the ‘science denier’ style labels got used and contrary views were considered ‘dangerous misinformation.’
Those were the gestalt of the era, not the words that the top experts were saying. Focusing in just on the first act:
Look at the words that Fauci or the CDC were saying in March 2020, and it wasn’t “wearing a mask won’t reduce your chances of getting covid at all.”
The quote from the CDC which got discussed on LW was “CDC does not recommend that people who are well wear a facemask to protect themselves from respiratory diseases, including COVID-19.” This isn’t directly saying anything about the world (except about what the CDC is recommending); the CDC generally speaks in imperatives rather than declaratives.
Fauci said sentences like “When you’re in the middle of an outbreak, wearing a mask might make people feel a little bit better and it might even block a droplet, but it’s not providing the perfect protection that people think that it is” (from this 90 sec video—IMO watching the video gives a better sense of what Fauci is up to). The content of that sentence is that the risk reduction from masks is probably > 0% but is < 100%, but it’s said in a way that can easily give an impression that rounds off to “don’t work.” Some interesting rhetorical moves here, like comparing to the standard of “perfect perfect”, and switching to very concrete language (“block a droplet”) and tentative phrasing (“might even”) when discussing the benefits rather than using any word like “effective” or “helps” or “prevents” which fits neatly into the concept templates that people are ready to act on.
It’s the sort of situation where, if you’re half paying attention to the chatter and inclined to believe what you hear, then you’ll come away thinking that masks don’t work. But my recollection is that, once I had reason to think that think mask-wearing would reduce my chances of getting covid (late Feb?), I didn’t come across any instances where a person who would know clearly said ‘no, you’re wrong about that, masks don’t work at all.’

Unnamed 18 Jan 2022 1:54 UTC
2 points
in reply to: Unnamed’s comment on: French long COVID study: Belief vs Infection
For comparison: imagine some medical researchers are interested in whether a particular medicine helps with a particular medical condition, so they set up a placebo controlled trial. A bunch of people with the medical condition all get their symptoms tested, then they flip a coin and half get pills with the medicine while the other half get sugar pills, and they don’t know whether they have the real pills. Then, some time later, they all get their symptoms tested again.
Now, imagine that I’m interested in “placebo effects”—I want to see if the ritual of taking sugar pills which you think might be medicine improves people’s health, or causes side effects, and I want to piggyback on this medical trial. I could just look at the pre vs post results for the set of people who got the sugar pills, but unfortunately this medical condition varies over time so I can’t disentangle effects of the pill-taking ritual from changes over time. I wish the study had a third “no-pill” group who (knowingly) didn’t get any treatment, in addition to the medical pill group and the inert pill group. Then I could just compare the results of the sugar pill group to the no pill group. But it doesn’t.
So I have the clever idea of getting the researchers to add a question to the tests at the end of the study, where they ask the patients whether they think they got the medicine pill or the sugar pill. That gives me a nice 2x2 design, where patients differ both in whether they got the medicine pill or the sugar pill, and separately in whether they believe they got the medicine pill or the sugar pill. So I can look separately at each of the 4 groups to see how much their condition improved, and what side effects they got. Changes that are associated with beliefs, I can claim, are based on the psychological effects of this pill taking ritual rather than the physiological effects of the substance they ingested.
This is a terrible study design. Who’s going to believe they got the real medicine? Well, people whose condition improved will tend to think they must’ve gotten the real medicine. And people who noticed physiological states like nausea or dry mouth will tend to think they’ve gotten the real medicine. This study design will say that improved condition & nausea are caused by people’s beliefs about whether they got the medicine, when in reality it’s the reverse: the beliefs are caused by these physical changes.
If I’m especially meddlesome, I might even tell the original researchers that they should use this 2x2 design to evaluate the original study. Instead of just comparing the outcomes for the medicine pill group and the sugar pill group, they should compare the outcomes while controlling for people’s beliefs about whether they got the medicine. That would mess up their study. It would be asking how effective the medicine is, after removing any effects that allowed patients to realize that they’d gotten the medicine (as if belief-entangled effects couldn’t be physiological effects of the substance).

Unnamed 18 Jan 2022 1:31 UTC
3 points
on: French long COVID study: Belief vs Infection
It’s tricky to run studies with beliefs as a variable, because beliefs have causes, so you’re setting yourself up to have confounds. I haven’t looked that closely at this study, but here are some possibilities:
Severity: people who had covid but believed that they didn’t had mild symptoms. So ‘severe cases have more long-term symptoms than mild/asymptomatic cases’ would look like ‘covid+belief leads to more reported long-term symptoms than covid without belief’.
Other illnesses: people who didn’t have covid but thought the did had some other illness like flu or pneumonia. If there’s long flu, then the long-term symptoms could be from that.
Long-term symptoms: a person who thinks that they probably just have a cold and not covid, but then is still fatigued a month later, might conclude that actually it probably was covid. So medium-to-long-term symptoms can cause belief, rather than belief causing long-term symptoms.
Testing inaccuracy: if the test that they’re using to establish the ground truth of whether a person had covid isn’t that accurate, then people who they’re counting as ‘covid but no belief’ might actually be false positives, and people who they’re counting as ‘no covid but yes belief’ might be false negatives.
Hypochondria: people who are to imagine that their health is worse than it actually is might mistakenly believe that they had covid (when they didn’t) and also imagine that they have long-term symptoms like fatigue or difficulty breathing. If people who did get covid have similar reported long-term symptoms, that means that the actual long-term symptoms of people who had covid are as bad as the imagined level of symptoms among people who imagined they had covid.
Denial: the reverse of hypochondria—people who say they’re fine even when they have some health symptoms might say that they didn’t have covid even though they did, and then downplay their long-term symptoms.
Trolling: if data slipped into the study from any people who find it funny to give extreme answers, they would look like hypochondriacs, claiming to have covid & long-term symptoms even if they didn’t have covid.
The first few of these possibilities are cases where facts about the world influence beliefs, and those facts also influence long-term symptoms. The last few of these possibilities are where the person’s traits influence their beliefs (or stated beliefs), and those traits also influence their reports of what long-term symptoms they’ve had.
If you wanted to independently assess the effects of getting covid and believing that you had covid, ideally (for scientific rigor unconstrained by ethics or practicality) you’d randomly assign some people to get covid or not and also randomly assign some people to believe they had covid or not (e.g. by lying to them). If you couldn’t have perfect random assignment + blinding, then you’d want to measure a whole bunch of other variables to account for them statistically. In reality, without anything like random assignment, who gets covid is maybe close enough to random for an observational study to work, especially if you control for some high-level variables like age. Beliefs about whether you had covid are heavily entangled with relevant stuff, in a way that makes it really hard to study them as an independent variable.
Is there good reason to think that this study overcomes these problems?