Link Search Menu Expand Document

Notes | Tutorial 36

Date: 2020-09-04

pre-tutorial - postmortem thoughts

  • I made a big tree on some thoughts on solomonoff induction the other day.
  • I’m not sure if I effectively reached my goal
    • the goal was to explore some things about the way SI views hypothesis programs
      • I think some things are left out which are important, like how details about how data is encoded. particularly stuff like parameters, e.g. the scenario params for how data was gathered.
      • Also I think there might be a way to find deficiencies of SI via self-referential problems. e.g. a program which generates its own parameters – note: this example isn’t complex enough to actually test, but I think some more sophisticated constructions might be able to.
    • Some problems I think I ran into:
      • most of the ~2000 words are not related to SI, but to setting up a conceptual model of how ideas work and what things we need to evaluate them
      • I got to the end and it wasn’t necessarily clear how to construct a self-referential problem, though I might just need to spend a bit more time thinking on it
      • I ran into limits of my knowledge on SI
        • in part this is due to low-quality accessible explanations of it, and inaccessible academic formulations


  • bottlenecks
    • e.g. AnneB watching video lectures
    • look for - high leverage points
    • try to give a little bit to get past stuck points
      • a bunch of stuff - most is stuff they could do on their own
      • if they couldn’t do it on their own it won’t help anyway
  • CR - multiple roads to truth, but not many ways known to build to an idea (like 1-3 ways); only a few foundations that (ppl think) matter
  • activity selection and time prioritisation important
    • be careful of recency bias
  • older stuff
    • feels like you’ve changed a lot
    • that stuff can jump out - so easy issues to find
    • ppl say “i’ll stay”, “i’m into truth seeking”
      • ET spots a problem
      • they deny
      • a way of evading/avoiding discussion about the big problem
    • ppl dismiss things as “small errors” - but you can’t measure that before you have a soln. knowledge (and reach) is unpredictable.
  • can do better with a good map of relevant field/knowledge
    • what builds on what
    • what are the open problems
    • are there areas you haven’t looked in to?
    • can’t be perfect but can do decent surveys

ideas for SI tree postmortem

  • did not establish if the foundation has significant errors
    • if it does the stuff I build off that foundation will have errors
    • idea: work more slowly, write posts/trees/whatever about the foundations first, then build on after EC / feedback / criticisms.
  • ET: jumping way too much into detail
    • generally better to start at higher level
      • main concepts, what problems are they trying to solve, if that’s sorted out then look at details of soln and whether to analyse some of the details for issues
      • a lot of the time one doesn’t get to detail level, things fail in earlier stages (e.g. overall consistently)
      • jumping into details common in convos, but more time on meta issues, organisation, where to focus => more efficient
  • max’s intuition on parallel constructions
    • 35 min in
    • trying to do bridging often not most efficient way to figure out issues; just looking at them on their own terms
    • bit of a last resort due to added complexity/difficulty
      • internal contradictions preferable
    • something internal has(?) to be wrong if idea is wrong
    • usually don’t have to get into rival frameworks
      • hard to deal with, usually not necessary
    • ET: this stuff too foreign to their system - what are their building blocks, the power it has, etc
      • important to try to understand ideas on their own terms before disagreeing or introducing own stuff
      • can focus on ‘what is this idea’ and try and understand it instead of distracting with diff system
      • if it’s hard to analyse their ideas and introducing stuff, ask if they have a way to do it before using your own ideas
      • criticising incompleteness often a lot simpler / easier than trying to introduce stuff
      • better if they introduce stuff than you (as like an outsider)
        • show problems, rather than telling them “imma add all this stuff”
        • will seem unfair otherwise
      • goldratt talks about similar stuff with e.g. how to do presentation; present problem and ask how other guy would solve it - esp if he can come up with soln himself, then on the same page
        • check “it’s not luck”, goldratt satellite program (50min)

other tutorial ideas

  • responding to something, e.g. melting democracy

outline from memory

  • some ppl like liquid democ
    • there are issues with direct democ
    • LD ppl are right to see those
    • LD solves some issues
    • but LD has issues too
  • there are good things about RD
    • like time allocation to understand issues - reps have that but it’s not practical in DD
    • also longer projects - reps allow ppl to do multi-year projects instead of worrying too much about being cut off tomorrow (which could happen in LD)
  • So I’d like to propose a watered-down version of liquid democracy

  • melting democ
    • there are aims of melting democ (note, this might be before MD in post) – 4 key points
      • (10/10) politicians run better campaigns
      • (5/10) better flow of information
      • (6/10) projects and specialisation easier than LD/DD
      • one more I can’t remember
    • maybe there is a good middle ground
    • we could do elections with LD
      • so ppl delegate, then we get like super delegates that choose reps
      • this would allow for some more efficiencies like SDelegates being able to talk to candidates and have high-yield conversations (as opposed to candidates talking to individual members of the public)
  • still some problems
    • need to wait till next general election (like 3 yrs)
    • super delegates might end up just showing problems that already exist (i.e. problem not solved)
      • A potential problem is that we’re centralising voting power in superdelegates, which might open the door to corruption.

  • some similarities to other stuff
    • like things in ireland or switzerland(?)
    • comment: feels like citizen juries and stuff
    • potential here

stuff I missed:

  • stuff about new parties / new ideas

if I were to reply

  • goals:
    • starting a dialogue - this guy might be useful to talk to
      • b/c he might advocate my stuff
      • might provide good criticisms of my ideas
        • different subculture
      • has been thinking about this and disagrees with some stuff I disagree with (but is common)
    • content generation
      • writing practice
      • ppl might be interested
      • gives a context for explaining IBDD (which I haven’t done since doing tutorials with ET)
      • appearance of good will via dialog
    • get advocates for flux / IBDD (also mentioned above)
    • practice understanding ppls ideas and responding efficiently
      • with (ideally) high output

goals aren’t that specific to him, so no need to invest too much into stuff specific to him.

what’s a plan to achieve those goals, what things need to be included in the post, how much effort to put in

  • multiple posts needed
    • a comment on his post
    • a more detailed post on my site - directly regarding his post
    • post about IBDD not in direct reference to MD
    • ??
  • map out his ideas, what are the nexus points (i.e. are there any criticisms or ideas that have particular reach in this context)
  • comment should address particular points, include only best points
  • note that I’m looking for discussion and criticism of my ideas, willing to discuss ideas I haven’t considered
  • post the content to personal + flux sites

next actions:

  • (immediate) map out his ideas
    • noting issues
  • (immediate) write comment w/ some ideas + asking for discussion etc + can do before linking my response post
  • if he’s interested in discussing
    • (immediate) write post responding to his post
    • ET: be careful b/c ppl often don’t live up to expectations, find ways that are okay if they don’t do anything
      • multi purpose actions, parochial posts are risky, kind of a failure
      • more general posts are less dependent on other person and their actions are a bonus
      • safer to be parochial the more you’ve talked with them
      • can filter out 90% of ppl - whether they claim to be serious/important/interested/etc
  • (immediate, do anyway) write post on IBDD

if you know what info you need, then specifically try to get that info - say what I’m looking for

et: this stuff, it’s a bit complex, you want to look a prev discussions that have worked, don’t overextend relative to those. what’s an MVP of a discussion. once you can get that working you can expand on it.

prev discussions

a lot of discussion can start good and break later, fail bigger picture goals but first 2/3 meet short term goals.

it’s easier to have discussions w/ ppl who have a lot of writing. you’re less reliant on them to tell you (you can look it up, or they might tell you), more flexibility on what you want to look into

the worse their archive the more you rely on them

et: a problem with asking ppl for discussion - different goals (1h44m)

value of discussion: get and give crit/feedback

books: not fully complete, need to fill in gaps; other ppl can get answers, hard to catch all your own errors

BoI msg

alice says:

  • 15% through BoI - started reading “a couple of days ago”
  • finding it interesting
  • i think i will def need to read it at least twice though, since I am waiting until I get to later chapters to see if my questions from earlier chapters are answered

  • so I will need to go back and read it to see if they were

  • or i could always write them down as I go, but I have not been doing that


  • should write down Qs at least
  • can ask me or at whether Qs are answered later – light & low-expectation discussion


  • problems with reading twice, can be good but ppl often don’t follow through
    • main problem with that method
  • depends on type of qs, some are immediate problems
    • other qs are tangential curiosities – book isn’t wrong if it doesn’t go back to it
  • i can sorta understand but think they’re wrong, what would be answer to this counter-argument
    • important to write this stuff down
  • some are half-baked ideas, who cares if they’re lost
  • can post Qs to:
    • can post notes too - allows others to give feedback
  • another thing besides Qs:
    • if there are deal breakers, SAVE THE QUOTE AT THE TIME, really important
    • can’t have good discussion without it
    • write it down RIGHT as they get to it

You can leave a comment anonymously. No sign up or login is required. Use a junk email if not your own; email is only for notifications—though, FYI, I will be able to see it.

Comments powered by Talkyard.