I think the meta-point is that these are great questions, and are reasons that going to a bar and chatting people up is actually a high skill activity with a fairly high bar (ahem) to clear. Most activities attended by most people most of the time are either gated behind such skill checks or invitation only. A recurring philosophy meetup carefully engineered to have no such bar is actually a super specific and rare thing. If you kept considering events, and then rejecting them based on the sorts of questions listed here (i.e. they seemed intimidating or hard) until you came across an event which raised no such issues, that seems like it should lead rather reliably to out of distribution results.
yeah, i might acclimatize to worse waters faster if i i’m not in the habit of rolling my own weekly rationality meetups which are precisely the social environment that i enjoy most.
but as mentioned in a prev comment, if the end result of that is me turning evil when i try to interact with normal folks, that seems worth trying to address.
(uh, flagging that i’m not entirely following the thesis of your comment, but i’m responding to the last line of it.)
class UnselectiveEvent(SelectiveEvent): def try_to_join(self, skill_q): self.members.append(skill_q) return True
events = [SelectiveEvent() for _ in range(99)] + [UnselectiveEvent()]
for _ in range(1000): society_member = random.random() while not random.choice(events).try_to_join(society_member): pass print(np.mean(events[-1].members)) # e.g. 0.13823472583179908 ```
I augmented the code a bit to get the mean and stddev of the SELECTIVE events to illustrate how far out of distribution the UNselective event would predictably be...
$ ./selection_events.py What skill profile over the SELECTIVE events? N = 98 // Stddev = 0.20449118563464222 // mean = 0.6569978446874036 What is the average skill in THE UNSELECTIVE event 0.1496967789384321
I think the meta-point is that these are great questions, and are reasons that going to a bar and chatting people up is actually a high skill activity with a fairly high bar (ahem) to clear. Most activities attended by most people most of the time are either gated behind such skill checks or invitation only. A recurring philosophy meetup carefully engineered to have no such bar is actually a super specific and rare thing. If you kept considering events, and then rejecting them based on the sorts of questions listed here (i.e. they seemed intimidating or hard) until you came across an event which raised no such issues, that seems like it should lead rather reliably to out of distribution results.
yeah, i might acclimatize to worse waters faster if i i’m not in the habit of rolling my own weekly rationality meetups which are precisely the social environment that i enjoy most.
but as mentioned in a prev comment, if the end result of that is me turning evil when i try to interact with normal folks, that seems worth trying to address.
(uh, flagging that i’m not entirely following the thesis of your comment, but i’m responding to the last line of it.)
Sorry, I am not the best at expressing myself clearly in prose. This is closer to what I was actually thinking, is it more helpful?
```
import random
import numpy as np
class SelectiveEvent:
def __init__(self):
self.members = []
self.skill_check = random.random()
def try_to_join(self, skill_q):
if skill_q > self.skill_check:
self.members.append(skill_q)
return True
return False
class UnselectiveEvent(SelectiveEvent):
def try_to_join(self, skill_q):
self.members.append(skill_q)
return True
events = [SelectiveEvent() for _ in range(99)] + [UnselectiveEvent()]
for _ in range(1000):
society_member = random.random()
while not random.choice(events).try_to_join(society_member):
pass
print(np.mean(events[-1].members))
# e.g. 0.13823472583179908
```
this is the most lesswrong thing i’ve ever seen. never change
I augmented the code a bit to get the mean and stddev of the SELECTIVE events to illustrate how far out of distribution the UNselective event would predictably be...
$ ./selection_events.py
What skill profile over the SELECTIVE events?
N = 98 // Stddev = 0.20449118563464222 // mean = 0.6569978446874036
What is the average skill in THE UNSELECTIVE event
0.1496967789384321
Two and a half standard deviations worse!