Community Archive

🧵 View Thread

🧵 Thread (19 tweets)

Placeholder
Hazard@natural_hazard• over 5 years ago

I figured it out The Courage to Be Disliked is about goodharting and overfitting It's take: try not to do either

41 4
2/20/2020
Placeholder
Hazard@natural_hazard• over 5 years ago
Replying to @natural_hazard

CTBD has a quick aside that sheds light on it's whole frame. p: don't rebuke or praise. It is through seeking praise that one develops the belief that they have no worth this clicks for me when I think of it in terms of wireheading and goodhart

19 0
2/20/2020
Placeholder
Hazard@natural_hazard• over 5 years ago
Replying to @natural_hazard

You want to do stuff that's "good" (not necessarily moral. think good music, good code, good style, etc). How do you figure out what is good? Good relates to other people, so you look to them. They will say if they thing your acts are good or bad.

15 0
2/20/2020
Placeholder
Hazard@natural_hazard• over 5 years ago
Replying to @natural_hazard

Spam this enough, and begin to believe "I got praised for this" and "it is good" are the same thing. One must imply the other But it doesn't. And the more you think praise defines goodness, the shittier job it does of defining goodness

20 1
2/20/2020
Placeholder
Hazard@natural_hazard• over 5 years ago
Replying to @natural_hazard

boom, adversarial goodhart. If other people know you are steering based on the praise, they can use that praise to manipulate you even if the manipulation isn't extreme, it breaks the connection between praise and the og "good" you were tracking https://t.co/sIsCEk3LDA

22 2
2/20/2020
Placeholder
Hazard@natural_hazard• over 5 years ago
Replying to @natural_hazard

Where I really got tripped up with this was domains where it seems like "good" is almost entirely defined by other people's opinions (most performing contexts, the show was a bust, because no one liked it) But the same dynamic still applies, it's just more subtle

17 1
2/20/2020
Placeholder
Hazard@natural_hazard• over 5 years ago
Replying to @natural_hazard

People can control when they give praise. People can't really control if they enjoy your show. The *real* data you want to train on is "did people enjoy this?" Problem is you don't have pure access to that. So you look to see if they praise you for the show.

20 2
2/20/2020
Placeholder
Hazard@natural_hazard• over 5 years ago
Replying to @natural_hazard

This is bad cuz you can *sorta* wirehead. Or more, pay attention to praise long enough, and you *rewire* yourself to care about praise, and not "was it *actually* good?" Self-fullfilling prophecy. For you, praise HAS become the thing you viscerally fell and care about

19 1
2/20/2020
Placeholder
Hazard@natural_hazard• over 5 years ago
Replying to @natural_hazard

Now you live in a narcissistic hellscape where it is literally the case that image (i.e what other people decide to praise you for) is the only thing you can care about. I think a key piece of fighting this is @visakanv "cultivate taste" https://t.co/2aqvVw3wcx

Placeholder
Visakan Veerasamy@visakanv• almost 8 years ago

I believe that taste is one of the most powerful, precious, fragile things in the known Universe and should be cultivated accordingly https://t.co/yt9Fv4lAsI

73 21
25 1
2/20/2020
Placeholder
Hazard@natural_hazard• over 5 years ago
Replying to @natural_hazard

@visakanv You NEED to develop your own sense of taste Depending on the domain, you will still need to pay attention to feedback from others, but somehow avoid treating it as the thing itself not sure what the mental moves of this look like, but just having the model probs helps

25 1
2/20/2020
Placeholder
Hazard@natural_hazard• over 5 years ago
Replying to @natural_hazard

@visakanv *scene change* Now we're onto Overfitting "You can't please everyone, and if you try, you'll displease everyone" heard this before but still secretly thought that I *could* get everyone to like me. CTBD helped me meditate on why you really just can't do this

14 1
2/21/2020
Placeholder
Hazard@natural_hazard• over 5 years ago
Replying to @natural_hazard

@visakanv A frame that I implicitly buy into all the time is: "If I was DOPE, people would like me, cuz how you can you not like something DOPE?" so when someone doesn't like me, it's tempting to figure out why and then "fix" the "problem"

16 2
2/21/2020
Placeholder
Hazard@natural_hazard• over 5 years ago
Replying to @natural_hazard

@visakanv This is a lie. Yes, there are things you can do that that stack the deck in your favor (be less judgy, don't be an asshole), but there IS NO SUCH THING as a "universally like-able person" THERE ISN'T THERE REALLY ISN'T

22 2
2/21/2020
Placeholder
Hazard@natural_hazard• over 5 years ago
Replying to @natural_hazard

@visakanv consequence: figuring out why [insert person] doesn't like you, and then "fixing" the problem behavior is not a process that pushes you towards Platonic Likableness it's just applying an haphazard patch, one that is bound to be at odds with one you apply for [other person]

11 1
2/21/2020
Placeholder
Hazard@natural_hazard• over 5 years ago
Replying to @natural_hazard

@visakanv AT BEST you devote massive attention and resources to analyzing the social setting of a given moment and mold yourself to the "ideal person" for that context, and do this over and over again, recalculating every time ANYTHING changes ..../

17 1
2/21/2020
Placeholder
Hazard@natural_hazard• over 5 years ago
Replying to @natural_hazard

@visakanv /...this is a HUGE drain on your soul/psyche/energy. Keep this up and it will consume so many resources that other parts of your life crumble the crumbling of these other parts fuck up your ability to have interactions with people, even you fake chameleon ones .../

16 1
2/21/2020
Placeholder
Hazard@natural_hazard• over 5 years ago
Replying to @natural_hazard

@visakanv /... oops, now you're unpleasant to be around and people don't like you. You sacrificed your personality to avoid being disliked, and ended up....... being disliked haha, this joke is funny

20 1
2/21/2020
Placeholder
Hazard@natural_hazard• over 5 years ago
Replying to @natural_hazard

@visakanv Contemplating how to burn this deeper into my bones... but yeah, "be liked by everyone" not possible if you're like "I just want to be liked by most people" you're probably falling for the same trap. if you're gonna chase being liked, narrow your scope https://t.co/lvGtJEX3yB

Placeholder
Visakan Veerasamy@visakanv• about 6 years ago

what's a minimum viable scene? if you have two sufficiently obsessive people who are trying to impress and outdo each other in public, two is enough but usually it seems that it takes a broader/wider scene to generate 2 such obsessive people https://t.co/N3qBkYWy12

Quoted tweet image 1
421 48
16 1
2/21/2020
Placeholder
lemon (is)@lemonaut1• over 3 years ago
Replying to @natural_hazard

@natural_hazard @visakanv This is a great point. In terms of cultivating your inner taste, it's helpful to STOP trying to be Universally Likeable, and instead perhaps weight heavily into a few trusted "taste-makers" who you admire/whose opinions you really care about.

2 0
5/29/2022