I’m glad you like it!
Fixed the footnotes. They were there at the end, but unlinked. Some mixup when switching between LW’s Markdown and Docs-style editor, most likely.
Pablo Repetto
Karma: 260
I see… so trolling by patenting something akin to convolutional neural networks wouldn’t work because you can’t tell what’s powering a service unless the company building it tells you.
Maybe something on the lines of “service that does automatic text translation” or “car that drives itself” (obviously not these, since a patent with so much prior art would never get granted) would be a thing that you could fight over?
You’re welcome! I’d like hearing a bit about how it helped, if you are ok with sharing.
Hi! I wrote a summary with some of my thoughts in this post as part of an ongoing effort to stop sucking at researching stuff. This article was a big help, thank you!
I’m glad you enjoyed it! I agree that more should be done. Just listing the specific search advice on the new table of contents would help a lot.
I’m gonna do the work, I promise. I’m just working up the nerve. Saying, in effect, “this experienced professional should have done his work better, let me show you how” is scary as balls.
First of all: thank you for setting up the problem, I had lots of fun!
This one reminded me a lot of D&D.Sci 1, in that the main difficulty I encountered was the curse of dimensionality. The space had lots of dimensions so I was data-starved when considering complex hypotheses (performance of individual decks, for instance). Contrast with Voyages of the Grey Swan, where the main difficulty is that broad chunks of the data are explicitly censored.
I also noticed that I’m getting less out of active competitions than I was from the archived posts. I’m so concerned with trying to win that I don’t write about and share my process, which I believe is a big mistake. Carefully composed posts have helped me get my ideas in order, and I think they were far more interesting to observers. So I’ll step back from active competitions for a bit. I’ll probably do the research summaries I promised, “Monster Carcass Auction”, “Earwax” (maybe?), then come back to active competitions.
Thank you for doing the work of correcting this usage; precision in language matters.
I made some progress (right in the nick of time) by...
Massaging the data into a table of every deck we’ve seen, and whether the deck won its match or lost it (the code is long and boring, so I’m skipping it here), then building the following machinery to quickly analyze restricted subsets of deck-space.
q = "1 <= dragon <= 6 and 1 <= lotus <= 6" display(decks.query(q).corr()["win"].drop("win").sort_values(ascending=False).plot.bar()) decks.query(q)["win"].agg(["mean", "sum", "count"])q is used to filter us down to decks that obey the constraint. We then check the correlation of each card to winrate. Finally, we show how many decks were kept, and what the winrate actually is.
q can be pretty complicated, with expressiveness limits defined by
pd.DataFrame.query. A few things that work:(angel + lotus) == 01 <= dragon and 1 <= lotus and 4 <= (dragon + lotus)1 <= dragon and lotus == 0(pirate-1) <= sword <= (pirate+1)
My deck submission (PvE and PvP) is:
4 angels 3 lotuses 3 pirates 2 sword
See response to Ben Pace for counterpoints.
My counterpoints, in broad order of importance:
If you lie to people, they should trust you less. Observing you lying should reduce their confidence in your statements. However, there is nothing in the fundamental rules of the universe that say that people notice when they are deceived, even after the fact, or that they will trust you less by any amount. Believing that lying, or even being caught lying, will result in total collapse of confidence without further justification is falling for the just-world fallacy.
If you saw a man lying to his child about the death of the family dog, you wouldn’t (hopefully) immediately refuse to ever have business dealings with such a deceptive, amoral individual. Believing that all lies are equivalent, or that lie frequency does not matter, is to fall for the fallacy of grey.
“Unethical” and “deceptive” are different. See hpmor ch51 for hpmor!Harry agreeing to lie for moral reasons. See also counterarguments to Kant’s Categorical Imperative (lying is always wrong, literally never lie).
The point about information theory stands.
Note that “importance” can be broadly construed as “relevance to the practical question of lying to actual people in real life”. This is why the information-theoretic argument ranks so low.
If good people were liars, that would render the words of good people meaningless as information-theoretic signals, and destroy the ability for good people to coordinate with others or among themselves.
My mental Harry is making a noise. It goes something like Pfwah! Interrogating him a bit more, he seems to think that this argument is a gross mischaracterization of the claims of information theory. If you mostly tell the truth, and people can tell this is the case, then your words convey information in the information-theoretic sense.
EDIT: Now I’m thinking about how to characterize “information” in problems where one agent is trying to deceive another. If A successfully deceives B, what is the “information gain” for B? He thinks he knows more about the world; does this mean that information gain cannot be measured from the inside?
Followed up on this idea and noticed that
A table of winrate as function of number of “evil” cards and “item” cards shows that item cards only benefit evil decks. I considered dragon, emperor, hooligan, minotaur, and pirate to be evil.
“No items, all good” wins 55.6%
“4 items, no good” wins 58.4%
“4+ items, all good” peaks at 37.0% winrate, and drops when adding items
Sorry in advance for an entirely too serious comment to a lighthearted post; it made me have thoughts I thought worth sharing. The whole “Karma convertibility” system is funny, but the irony feels slightly off. Society (vague term alert!) does in fact reward popular content with money. Goodhart’s law is not “monetizing an economy instantly crashes it”. My objections to Karma convertibility, are:
Exploitability. Put in enough people on a system, and some of them will try to break it. Put in even more people, and they will likely succeed.
Inefficiency. There is a big upfront cost to generating good ideas. Ie: you’ve got to invest putting interesting things inside your mind before you can make up your own. Specialization pays off just as much in intellectual labor as in other areas.
Pareto’s law. A small, selected group can be a consistent performance outlier.
Marginal returns on money. The marginal value of money goes down dramatically when you get above the subsistence threshold.
I think that providing a salaried position to a small number of intrinsically motivated individuals is a much more efficient way of buying ideas. I think RAND is basically structured this way?
Epistemic status: Don’t quote me on any of this. I’ve done no research, instead I’m pattern-matching from stuff that I’ve passively absorbed.
If there is a place where “pay no attention to that man behind the curtain” is to be disobeyed with prejudice, I’d reckon this is it. ;)
Just the obvious contrarian poke:
Are the decks equally likely? We observe that 412050 decks appear just once, 104483 decks appear twice, etc. Is this distribution compatible with random draws?
There are 342396 rows, with 2 decks each. Solving for the number of valid decks one could make, gives me (straightforward application of counting particle arrangements, imagine you have “coins” to place in “card-type boxes”).
Then I just simulated and eyeballed. If I pick 2*342396 random numbers from 1 to 1352078, how many numbers appear just once? How many twice? Eleven runs and the data looks like this:
observation [412050, 104483, 17805, 2257, 246, 16, 1] simulations [411712, 104617, 17885, 2245, 213, 23, 1] [412504, 104559, 17665, 2240, 212, 20, 5] [412884, 104509, 17528, 2256, 225, 25, 1] [413170, 104376, 17629, 2190, 217, 23] [411926, 104692, 17828, 2209, 202, 23, 2] [411883, 104746, 17624, 2302, 244, 16, 3] [411866, 104728, 17779, 2258, 190, 24, 1] [412771, 104106, 17813, 2294, 210, 24] [412746, 104561, 17640, 2205, 209, 22, 1] [412083, 104869, 17627, 2205, 237, 13, 1] [412093, 104377, 17854, 2287, 224, 18, 1]Which seems compatible enough.
I’m interested! Thank you for the heads-up. I’ll be sure to pop by.
Hm. I expected to do terribly on this problem since I hardly exhausted my avenues of research (I didn’t even clean up the infohazardous object, despite knowing about it from other’s comments). I ended up doing the worst out of everybody, which tracks, but the results are still clustered rather close.
I enjoyed the problem a lot, and I’m very grateful to aphyer for pinging me when he made the problem available. Tragically, I was sick at the time. :P
Brief notes:
Unlike Umberto Eco’s “How to Write a Thesis”, and like johnwensworth’s “How To Write Quickly While Maintaining Epistemic Rigor”, this does not propose you sacrifice the importance of questions on the altar of rigor
There is a pretty big hole on how to find “the 1-3 most prominent pieces on each side”. When first encountering a claim on the context of an academic debate, this is usually not an issue. Often, not so much. Obviously, if you simply came up with the claim you have no conveniently laid out pointer to the academic literature.
“Just write something”, or the vomit pass, is an invaluable technique. I have been doing this more and more over the last two years. Having full license to be wrong is incredibly freeing, and it makes it much easier to iterate on an idea. You just read what you wrote and pay attention to the parts that make you cringe.
I came in pretty late, so I haven’t gotten much to share.
I split my analysis into expected-profit-if-obtained and chance-of-obtaining-per-team. It probably isn’t true, but assuming that team selection does not directly affect profit simplifies things a lot.
Predators incur losses, except Euclid humanoids which are profitable.
Safe objects are profitable.
Euclid objects are profitable, except locations and replicators which are neutral.
Keter objects incur losses, except humanoid organics which are profitable.
When objects we obtained in the past are split by this criterion, “unprofitable objects” incur a mean loss of 25.5 (stderr 1.85), “neutral objects” make a mean profit of 5.4 (stderr 2.23), and “profitable objects” make a mean profit of 22.6 (stderr 0.86)
As abstractapplic pointed out, infiltrations do well on sites 2 and 6, legal teams on sites 4 and 7, and paramilitary teams on sites 3 and 8.
Paramilitary teams never get virtual targets, but do well on organics.
Infiltration teams do well on predatory and virtual targets (in particular, they do better than paramilitary teams against predatory Euclid humanoids).
Final allocations
Infiltration: SCP-4370, SCP-4390, SCP-537
Legal: SCP-3212, SCP-3597, SCP-1282
Paramilitary: SCP-3339, SCP-3440, SCP-5136
Early COVID response on LW was a generalized “this is a big deal.” I can’t find the post that originally caught my eye, but I remember hitting the supermarkets in Buenos Aires, stocking up on masks and hand sanitizer, and two weeks later seeing the city freak the hell out. Jacob’s “Seeing the Smoke” was a strong early signal, and Zvi’s updates often considered explicit numbers.