Posts
Comments
Your criticisms are mostly correct. I wrote the post to justify my actions rather than tell robust truth. Posting it as-is on LessWrong was my mistake.
"Entangled closer with physical reality" was a poor choice of words. I meant something closer to "experience my surroundings in more detail".
Reducing what you need implies broadening what you tolerate, in the same sense that a system with fewer axioms has more models. Interpreting it as twisted greed-avoidance is novel and odd to me. If you get used to walking barefoot, then you can better handle situations where you lack shoes. On further reflection, that broadening is small compared to other methods (as learning a language).
What exactly does "predictable" mean here?
You can infer the toki pona word (phrase) to match a meaning by joining words (standard base concepts) according to meaning-clusters of the base words and rules for adjective order. That is, making a toki pona word-phrase, you only need to understand the intended meaning of the whole phrase and the small set of base words.
Likewise, understanding a word-phrase to a good approximation depends only on the words in it and their arrangement. Understanding it exactly depends on context and conventions that build up around common terms.
If the phrase for "phone" means "speech tool", how do I tell between phone and loudspeaker or cough drop?
You can add more adjectives ("phone" could be "tool of distant speech" and "loudspeaker", "tool of strong speech"), or cope via context.
If I want to say "apricot" do I need to say "small soft orange when ripe nonfuzzy stone deciduous tree fruit"? Or do I just say something shorter like 'orange fruit' and hope the other guy guesses which kind of orange fruit I mean?
The latter is exactly what you do. If context leaves ambiguity, you add as many adjectives as needed, changing "fruit" to "orange fruit" to "small soft orange stone tree fruit".
How would I say "feldspar"? "Rock type #309"? How would I say "acetaminophen"?
Toki pona is less opportune when you need great precision like that. I see three solutions
- mash together lots of adjectives (feldspar = silicon-oxygen crystal + other details = square rock of bodily air and of moderate power movement ...)
- use numbers and symbols according to reductionism and the topic in question (acetaminophen = one-circly two-armed "C8H9NO2")
- bring in a loanword/proper adjective ("misikeke Asitaminopen")
If you call a multi-word phrase a word, we can more appositely claim that the formation of words and their associations to meanings, in toki pona, is very systematic and predictable. However many words it truly has, toki pona remains very easy to learn. The definition of "word" is flexible/arbitrary, but that final observation is most obviously consistent with the few-words view.
- You would wash your hands properly at all the appropriate times.
- You would study with spaced repetition.
- You would stop looking at (mainstream, megacorporate) social media.
The ability to quickly recall what I studied for its application.
I thought that was obvious. Why do you ask? What am I missing?
You almost always have some information to concentrate your priors. Between mutually-helpful speakers, implicit with an answer to a question is that the answer gives all the information you have on the question that could benefit the questioner. E.g.
What will the closing price of Apple be at the end of the year?
"Almost certainly somewhere between $150 and $250."
positive statements like "Stay away from the wires" are more effective than negative statements, like "Don't touch the wires," because your brain basically ignores the negative part of it. "*mumble mumble* touch the wires? Don't mind if I do!"
That's what I was going for with
When reading or hearing a negation used in language, you must first process the positive form it contains to understand the entire statement. For example, to understand "the sky is not green", you must first understand "the sky is green", then negate it. Usually, this happens quickly and subconsciously, but it can harmfully slow down or weaken understanding by making you first consider a false idea.
I predict that it mostly gets worked around, by using only a few extra words.
"The sky is something other than blue" and "I will be somewhere else tomorrow" are both semantically-equivalent to the forbidden forms. Even "I deny that the sky is blue" is a positive-form negation of the object-level statement.
I suspect all such workarounds depend on one of a relatively small set of negation-enabling words, such as "other", "else", and "deny", as you demonstrate. Prohibiting more words should eventually block all workarounds, while making writing more annoying.
An excellent alternative. I was going for something usable without any tools.
- beat up the lock by ordinary methods
- contact someone outside to let you out
- beat up the door by ordinary methods
- beat up the wall by ordinary methods
- teleport
- contact someone outside to destroy part of the wall
- reshape phone and/or clothes into paperclips and wait for the paperclip maximiser to take them in a way that will probably let you out
- break off a sharp piece of metal from your phone and cut your way out
- break off a thin piece of metal from your phone and pick the lock
- wish/pray
- order a delivery, which will require the door to be opened (during which you can walk out), which may be easier from the outside
- look up solutions and implement them (or contact person or AI to come up with them)
- ignite a fire (friction or something, perhaps aided by starting on clothes you take off), and burn away the wall
- distract yourself with entertainment on your phone so you mentally escape
- wear steel-toed shoes as part of your clothes and kick holes in the wall
- mentally swap inside and outside of room (the wall separates the two, but it doesn't care which direction) -- now you're outside
- wait for someone to check on you and let you out
- scream for help (or play suitably loud sound on your phone) to accelerate 17
- vaporise yourself (never mind how) and let your molecules diffuse to the outside world
- order a very thin, flat sheet of powerful explosives, which can be delivered thru a gap in the door without outsiders having to unlock, and detonate it to break the wall
- don't go in the room in the first place
- order a time machine and go back to prevent the door from being locked
- collapse to a true vacuum (... which would also kill you)
- look for mechanical weaknesses in the architecture of the room and exploit them
- write out enough of your thoughts to your phone to effectively transfer your mind/personality, then send it to be simulated outside the room
- hack the room's lights to be way brighter and gradually apply light pressure to push away part of the floor (you have 10 years, right?)
- use the other door which isn't locked
- order a saw and cut the wall easily
- scream at the wall (or play suitably loud sound on your phone) to break it
- open (or break, if necessary) the window and climb out
- publish a prize/bounty offer online for letting you out and wait for someone to take it
- wear a cape as part of your clothes and write (lay out thread pulled from other clothes?) in Spanish on it, making it an es-cape (... sorry)
- corrode the lock (or otherwise cause escape-enabling damage) with urine
- fall asleep and forget about your problem, another form of mental escape
- run really fast to slam the door open despite the lock when you hit it
- wait, one of the walls was missing this whole time?
- the lock's electronic, hack it
- break the light fixture to get glass shards from the bulb, which would be very effective for cutting your way out
- become the type of person who would take brutal revenge on those who lock you in rooms, thereby acausally preventing this situation
- take off some clothes, thread it around the door, get it stuck with part sticking out on your side, and pull really hard (and hope that overrides the lock)
- cry me a river (literally), which might break something by water pressure
- suggest a study of this room's lock to LockPickingLawyer
- disable phone's cooling mechanisms, heat the processor, melt a hole in the wall
- wait, the door didn't properly block the way out this whole time?
- talk to the gatekeeper over the text channel, be really smart and manipulative, and convince them to let you out
- set some clothes on fire, push it thru the gaps around the door, thereby making everyone else in the building panic and let you out as part of the emergency response
- skew the reviews of the company that owns the room to alter the number of people who come near the room, making someone let you out sooner (this would intuitively benefit from more people, but maybe less, sith bystander effect)
- you didn't say what kind of phone; maybe this is a special phone with an extra feature specifically designed to let me out
- wait, the lock was trivially insecure the whole time?
- go to the moon, which is outside the room (we have hundreds of ways of doing that)
Good challenge. I thought I could do this quickly, but it took ~45 minutes.
I looked things up when clarifying/writing out answers, but not in coming up with them.
Some of my answers are indirect, with the assumed completion "and then it'll be much easier to find an actual method". Some (a bigger some) are stupid.
Any otherwise-unclarified mention of "it" refers to the object which we want to send to the moon. Any otherwise-unclarified implicit reference to a task/goal refers to this task of sending it to the moon.
1 to 10: rocket, space elevator, catapult, make Earth bigger, bring moon closer, kick it really hard, outsource task to [other person/agent], build an unreasonably tall skyscraper, drive it there (find a way to grip tires onto near-vacuum?), direction-selective gravity disable + walking
11 to 20: set off explosives under it, machine-teleport, be on the moon already and pull it on a rope, motivate it to get away itself (if it's intelligent/independent), pray really hard, trebuchet, pump it up a pipe, move Earth towards moon, destroy Earth so rocketry etc is easier, swim
21 to 30: figure out something with graphene (it does enough other crazy things), wind-up spinny launch thing, booming force of loud noise, figure out where someone with name "Moon" lives and use conventional Earth travel, transfigure with tension, revolve it on a flywheel and release, send it into orbit so it is the moon, jump to an alternate universe where it's already on the moon, shoot matter against it to push it away, break apart the moon so its fragments land on Earth and you can access them by conventional travel
31 to 40: vaporise the moon so anywhere is technically "on all zero the-moons", make a knowingly-perversely-incentivised prediction market, break it to atoms which will all eventually diffuse to the moon (but probably not at the same time), quantum-teleport, let a nearby Earth-disaster blast it away to the moon, get people to collaboratively throw it, burn it (??? cf atomic diffusion maybe???), set a bounty, stretch it out by a factor of many thousands, plant something that grows indefinitely under it
41 to 50: stretch Earth, corrugate spacetime (yeah idk; ask my Babble), throw a rope to connect to the moon and climb, pile up a sufficiently-large mountain under it, climb a ladder, wait for red-giant Sun to make things "convenient", get a machine that'll turn it (and only it) into paperclips on the moon, establish a trend of sending Earthly things to the moon which will include it eventually, contrive your incentives/reward/emotion/utility function to want it harder, inflate a bubble under it
Just did it. Paradox.
Clever, but
to the point that I can't predict it
not further. If you increase redundancy, still unpredictable, as here, you probably went too far.
是年龄的影响，还是标识的影响？
机译
1 and 2 are absolutely correct, but for specific subsets. Outside such subsets, this optimisation still applies.
3 is correct sometimes as reversed advice. I see your point in 3 often (usually implicit). My post reverses that in response to it sometimes going too far.
It seems I went too far. Hence the expanded original:
Adjust how much to omit based on the concentration and domain-intelligence of the listener. Your starting point should probably err more on the side of "omit more redundancy" than it currently does.
Neither. Long-lasting deliberate idiosyncrasy, based on Shakespearean English.
What word is sufficiently Levenshtein-close to "sith" as to get there from a typo whilst also fitting grammatically into the sentence?
When you actually do these never-ending simple tasks, do you dislike/suffer-from the process itself? Or is this just the stress of having to do them sometime, when you're not doing them?
(Sorry if you already explained this, but it's not very clear from the question)
What did you think the right word would be?
(It's deliberate. Synonym of "because".)
I intended the latter. Ideally, instructors would start teaching students that in an act of educational reform, but that's harder and very unlikely from what I see now.
there are no widely-accepted models of how history works that are detailed enough to let you predict the outcomes of unfamiliar historical events
We don't need such a model. The students would be figuring it out for themselves, and we don't expect them to predict in great detail. There'd have to be a lot of partial credit in this.
in the few cases where students are asked to give causal explanations in current classes, their work is graded as a persuasive essay rather than as a factual claim that can be held to some objective standard of correctness.
That's exactly the issue.
Thanks, but that's deliberate. Revived-archaic synonym of "because".
Nice. Is there some post about dath ilan that establishes that (I don't see it in Yudkowsky's original AMA), or did you just make this up?
I don't understand what you're getting at with your response to the question of personal evil.
You're right about trading.
Seeing an Onion headline say "X did Y" is teeny evidence that X didn't Y.
...
I think that's doubtful.
In which direction? Do you mean to say that it's no evidence, or it's strong evidence? You speak of "a strong prior against the event", but the strength of the prior doesn't have any necessary relation to the strength of the evidence.
Fixed, thanks. I implicitly assumed that all ChatGPT use we cared about was about complicated, confusing topics, where "correct" would be little evidence.