Babble & Prune Thoughts

This is an ac­count­ing of var­i­ous thoughts I had when read­ing the Bab­ble & Prune se­quence.

En­courag­ing Babble

It is ironic, to me, that the Bab­ble & Prune se­quence ends with a call to ex­clude epistemic sta­tus tags from posts:

What hap­pens to the reader when ev­ery post starts, “epistemic sta­tus: mostly true with a chance of rain?”

[...]

In­stead of con­di­tion­ing read­ers to hate us, I pro­pose we re­turn to a saner time, where the fact that your words are but a pale wa­ver­ing shadow of the grand, mys­te­ri­ous truth in your heart is the de­fault as­sump­tion about hu­man com­mu­ni­ca­tion. Where truth is a dance of suc­ces­sive ap­prox­i­ma­tions yet no step in that dance re­quires adult su­per­vi­sion. Where quib­bles over cer­ti­tude are ban­ished to the com­ments sec­tion where they be­long.

I in­ter­pret Alk­jash as think­ing that, if the stan­dards were just lower in the first place, peo­ple would feel free to write more, which would get us fur­ther in the end (be­cause the good stuff can bub­ble to the top).

In­deed, my im­pres­sion is that the LessWrong team has worked to make LessWrong a space where peo­ple feel they can share raw thoughts rather than re­quiring ev­ery­thing to be care­fully re­fined.

But in my opinion, the “epistemic sta­tus” flags, and similar tools, help rather than hurt.

The New Twit­ter Ac­count Problem

There is a phe­nomenon—let me know if you think of a bet­ter name—which I call the new twit­ter ac­count prob­lem. My per­sonal in­ter­ac­tion with Twit­ter has been as fol­lows:

  1. Start a new twit­ter ac­count. Fol­low a few peo­ple I like. Ba­si­cally no fol­low­ers. No pres­sure! I can say what­ever I want! Dump some ran­dom thoughts into the new twit­ter ac­count over the next few days.

  2. Now I’m get­ting fol­lows, likes, and shares. Pos­i­tive re­in­force­ment! This en­courages me to keep it up.

  3. Fol­lows slowly climb, and I’m get­ting more and more in­ter­ac­tion. Now I feel like there’s some­thing at stake, and I have to do my best. I write some good tweets, but I grad­u­ally tweet less and less.

  4. I make an alt ac­count for “worse thoughts”, start­ing the cy­cle back at step 1. Grad­u­ally the alt ac­count be­comes my main ac­count, and I have the same prob­lem again, and need a third ac­count, and a fourth… Mean­while, the fol­low­ers of my pre­vi­ous ac­counts won­der why I’m not ac­tive any­more.

LessWrong 1.0 had a similar prob­lem. The front­page got “too good” for peo­ple to feel like they could re­ally post on. The “dis­cus­sion post” was in­vented. A lot of ac­tivity moved to dis­cus­sion. But then Dis­cus­sion got too good, and peo­ple felt like they had to have some­thing good even to post it in Dis­cus­sion. Open Threads were cre­ated within Dis­cus­sion, as the new low-bar-to-en­try fo­rum.

I’ve heard that a similar pat­tern has also played out on other dis­cus­sion plat­forms. In­creas­ing lay­ers of “no re­ally, it’s OK to post raw thoughts” are cre­ated as old lay­ers get too re­spectable.

This is prob­a­bly crazy, but...

In in-per­son dis­cus­sions, a good way to lessen this prob­lem is to pref­ace state­ments with qual­ifiers like “This is a dumb ques­tion, but...” or “Here’s a crazy idea:” or “I don’t en­dorse this, but I was think­ing...” and so on. This cre­ates a kind of pro­tec­tion for the speaker. They still get credit for good ideas; but if the idea re­ally was bad, they take less of a hit.

This doesn’t work as well for longform text dis­cus­sions, be­cause read­ers know you had time to think things through. But [epistemic sta­tus:] flags can play a similar role. If you don’t feel like an idea is good enough, you pref­ace it with [epistemic sta­tus: su­per dumb] or what­ever, and take some com­fort in the fact that if the reader doesn’t like what you wrote, they were warned.

I think hav­ing this norm is much bet­ter than at­tempt­ing to sus­tain a norm that ev­ery­thing is “epistemic sta­tus: raw thoughts” by de­fault.

Say Ran­dom Things

OK, so epistemic sta­tus flags are one way to com­bat the new-twit­ter-ac­count prob­lem. Do we have any other tools?

One tool is to pur­pose­fully lower your stan­dards. I be­lieve the book IMPRO in­cludes an ex­er­cise in which you point to ran­dom things around you and call them ab­solutely wrong things; (point to tree) “there’s a lamp-post” (point to grass) “there’s some spaghetti” etc. Per­haps say­ing ab­surd things on pur­pose helps prove to your s1 that noth­ing hor­rible will hap­pen if you say some­thing wrong. Another ex­am­ple of this is Allie Brosh drunk post­ing (no­table be­cause the ex­plains the thought pro­cess be­hind it). This is com­monly called shit­post­ing. Un­for­tu­nately, Allie Brosh wrote less af­ter that ex­per­i­ment, so it’s not clear that it had a pos­i­tive im­pact. Anec­do­tally, I’ve heard that a friend had dra­mat­i­cally pos­i­tive re­sults with the IMPRO ver­sion.

Master All Levels

I don’t think we should just work on im­prov­ing our bab­ble, though. I think it’s re­ally im­por­tant to as­pire to higher and higher stan­dards. I want to be­come stronger! How can we rec­on­cile this with the damp­en­ing effect high stan­dards can cre­ate?

Meta-Perfectionism

The true an­swer is that if “hold­ing your­self to a high stan­dard” makes you do/​say too lit­tle, then your con­cept of “stan­dard” is bro­ken. We in­tu­itively rea­son based on blame/​guilt, which makes im­proper in­ac­tion seem less bad than im­proper ac­tion. If we could free our­selves of that mode of rea­son­ing, per­haps we could just not have the new-twit­ter-ac­count prob­lem in the first place.

In The best you can, Nate Soares writes:

It’s easy to par­a­lyze your­self if you try to do the “right thing.” There’s always more un­cer­tainty to be had. There’s always more in­for­ma­tion you could gather. It’s hard to be­come con­fi­dent that you’re do­ing the right thing. This can lead to paral­y­sis, and per­sis­tent in­ac­tion.

It’s much eas­ier, I think, to stop ask­ing “is this ac­tion the right ac­tion to take?” and in­stead ask “what’s the best ac­tion I can iden­tify at the mo­ment?”

Some­times, the best ac­tion you can iden­tify is “search for more al­ter­na­tives.” Some­times, it’s “study more” or “learn more.” Some­times, it’s a spe­cific ac­tion. The nice thing is that “what’s the best ac­tion I can find in the next five min­utes?” always has a con­crete an­swer. If you search for that, in­stead, you won’t get par­a­lyzed.

He elab­o­rates more on similar thoughts in De­liber­ate once. But more rele­vantly to Bab­ble&Prune, his post Half-ass­ing it with ev­ery­thing you’ve got de­scribes the mind­set re­quired to ori­ent perfec­tion­ism at the meta-prob­lem of avoid­ing over­much perfec­tion­ism.

Max­i­mum Pay­off Per Effort

We face two prob­lems:

  1. In­vest­ing the min­i­mal amount of time/​effort to get the re­sults we want.

  2. Get­ting the best re­sults we can given the time/​effort.

Nate fo­cuses on the first prob­lem, de­scribing how some situ­a­tions call for a mea­sured effort, while other situ­a­tions call for an all-out effort. But we also want to do the sec­ond: the more fa­vor­able our pay­off per time/​effort spent, the less we need to spend on things call­ing for a mea­sured effort, and the more re­sults we get when we go all-out.

One thing uni­ver­sity art classes do is force you to spend a long time on a sin­gle draw­ing. This teaches you how good you can get if you spend a lot of time on a sin­gle piece. It also teaches you how to fruit­fully spend that much time on a piece.

But to grow as an artist, you also need to learn speed. You shouldn’t care­fully plan ev­ery draw­ing you do. Artists also prac­tice ges­ture draw­ings, in which you have a very short amount of time to cap­ture the gen­eral pose of a model. So, a de­liber­ate prac­tice of art in­volves prac­tic­ing at all time-scales.

There’s prob­a­bly a nat­u­ral slow-to-fast pro­gres­sion for many skills, where you need to go slow at first to be able to do it well, but can do it faster and faster af­ter that.

Train­ing Bab­ble, Train­ing Prune

Your prune should not just ac­cept/​re­ject. It should have de­grees. You should be able to step it up or down ap­pro­pri­ately. Fur­ther­more, you want your prune to give you spe­cific feed­back—not just “that’s bad”, but “that’s mis­matched”, etc. I sug­gest listen­ing to your in­ter­nal critic with a felt sense /​ fo­cus­ing lens. You may find that you know more than you thought about what makes for good work.

You want your bab­ble to be able to con­form to the high­est stan­dards it can while re­main­ing cre­ative. You want it to be like GPT-3, not en­tirely ran­dom words. So pay at­ten­tion to what your taste says about your bab­ble—the trick isn’t to cast your stan­dards aside, but rather to pay at­ten­tion to them with­out let­ting them stop bab­ble. Give your bab­ble a lit­tle breath­ing room from your prune.

Alk­jash men­tioned that bab­ble is not gen­er­at­ing things with in­de­pen­dent ran­dom­ness, but rather, is more like a ran­dom walk in con­cept-space. A lot of the skill of good bab­ble is in gen­er­at­ing good mu­ta­tions of ideas, not just good ideas. You can see this as you prac­tice bab­ble on longer time-scales. Give your­self more time to write a sen­tence, and you may see your­self go through sev­eral mu­ta­tions of that sen­tence be­fore set­tling on one to write. Part of what it means to spend more time on a draw­ing or paint­ing is giv­ing your­self more time to plan each line, and more time to fid­dle with it, seek­ing im­prove­ments.

My main point here is that im­prov­ing bab­ble doesn’t mean re­duc­ing prune. Alk­jash some­times speaks as if it’s just a mat­ter of open­ing the floodgates. Some­times peo­ple do need to just re­lax, turn off their prune, and open the floodgates. But if you try to do this in gen­eral, you might have ini­tial suc­cess but then ex­pe­rience back­lash, since you may have failed to ad­dress the un­der­ly­ing rea­sons why you had closed the gates to be­gin with.

The Many Gates

I think one of the most use­ful mod­els in Bab­ble & Prune was the three gates.

Ac­tu­ally, af­ter start­ing this sec­tion I re­mem­bered that I already wrote my thoughts on this in the Bab­ble & Prune sec­tion of Cap­tur­ing Ideas, par­tic­u­larly the part about de­vel­op­ing ideas. Go read that if you want a fur­ther elab­o­ra­tion of why just open­ing the floodgates isn’t ex­actly the goal.