the gears to ascenscion

Karma: 1,325

I read/watch lots of stuff shallowly, then share it; if you can do better with my links, open them and use them to seed your better lit review. self-taught, I’m often missing basic concepts; but, I do not defer on timelines. lately i’ve been dumbass 2 out of 8 in the comments often. speed up safe capabilities, slow down unsafe capabilities; don’t slow down safe capabilities, wtf! some current favorite links:

about me:

self-funded for a few more months. willing to trial for free at any org during that time; seeking $2.3k/mo-after-tax to do alignment research, or any other job; key criteria if job: it needs to be up to the task of syncing level of requested action from me to the level of agency my burnt-out adhd brain can output, with iterative increase as I restore capabilities. I’m a 10 year professional software engineer and this is an intentional salary cut for me calculated as living cost plus. after that I’ll probably get some crappy job doing who knows what. maybe food service, I swear I’ve lost iq.
ex startup founder. it went ok, not a unicorn, I burned out in 2019. couple of jobs since, quit last one early 2022.
lots of links in my shortform to youtube channels I like
would love to meet in vrchat, dm for username; I go to the transhumanist meetups fairly often, come say hi!
disagrees with many things.
doesn’t know enough math, what’s a sheave?

make all safe faster. end bit rot. forget no non-totalizing aesthetic’s soul.

(also, I type primarily with talon voice recognition, patreon-funded freeware which I love and recommend for voice coding; while it’s quite good, apologies for trivial typos!)

the gears to ascenscion 31 Jan 2023 22:35 UTC
2 points
0 ∶ 0
in reply to: Duncan_Sabien’s comment on: Basics of Rationalist Discourse
reasonable.

the gears to ascenscion 31 Jan 2023 22:02 UTC
2 points
0 ∶ 0
in reply to: porby’s comment on: Against Boltzmann mesaoptimizers
It’s probably fine-ish to allocate another reference to the concept, though I personally might suggest expanding it all the way out to “boltzmann brain mesaoptimizer”.

Are you familiar with restricted boltzmann machines? I think Hinton has described them as the other branch besides backprop that actually works, though I’m not finding the citation for that claim right now. In any case, they’re a major thread in machine learning research, and are what machine learning researchers will think of first. That said, boltzmann brains have a wikipedia page which does not mention lesswrong; I don’t think they’re a lesswrong-specific concept in any way.

the gears to ascenscion 31 Jan 2023 21:39 UTC
2 points
0 ∶ 0
in reply to: Duncan_Sabien’s comment on: Basics of Rationalist Discourse
I do not at all mean to criticize you deeply. this is a great post. I just want to be able to use it in conversations on discord where people are new to the concept with somewhat less difficulty. I linked it somewhere and got the immediate response “it opens with a quote from that one lady, close”, and another who was approximately like “geez that’s long, can you summarize”. Yes, I know you’d wish that the sanity waterline was higher than that; and you did do a great job building this ladder to dip into the sanity so the sanity can climb the ladder. I just wanted to have a link that would clearly signal “you don’t have to read the rest if you decide the intro isn’t worth it”. It’s a small edit, and your existing work into making the thing doesn’t make it impossible to change it further. honestly when I first posted my comment I thought I was being constructive and friendly.

the gears to ascenscion 30 Jan 2023 23:52 UTC
6 points
0 ∶ 0
in reply to: Charlie Steiner’s comment on: Call for submissions: “(In)human Values and Artificial Agency”, ALIFE 2023
no insider knowledge; I have some outsider knowledge that makes me optimistic that there will be interesting things to be learned from people there. alife research has been focusing on things like https://chakazul.github.io/lenia.html, and now particle lenia, https://google-research.github.io/self-organising-systems/particle-lenia/, and it connects well with other research on related topics that I’ve had a hobby of making indexes of, eg this recent post.

here are the proceedings of the 2022 conference: https://direct.mit.edu/isal/isal/volume/33 - some papers that seem worth a skim, from those proceedings (I really wish lesswrong had hover-to-preview, use a chrome tab group or something to open the links quickly then flip through them fast):
but as a person who’s always focused on approximations and is kind of beginner at actually doing the math I like to read about, I can only offer a pointer to the conference.

the gears to ascenscion 30 Jan 2023 19:28 UTC
10 points
0 ∶ 0
on: Tools for finding information on the internet
Great suggestions. I’d also add
for academics, https://arxivxplorer.com/ is a very solid semantic search engine for arxiv, beats semanticscholar for search; different strengths than metaphor and sometimes still fails to find stuff I know exists.
for academics, https://www.semanticscholar.org/ has a paper recommender, which you use by adding papers to folders and marking them as “give me a feed, please”; it does a great job finding stuff on topics I’m interested in as it comes out. Note that semanticscholar’s search leaves much to be desired, their model is best used in recommender mode.
for academics, https://my.paperscape.org/ is mostly not a search engine, but is convenient for mapping out which papers reference which other papers more quickly, and goes together nicely with other items on this list.
---
For general users, https://consensus.app/ is a convenient but limited question-answer search for papers. not as strong as elicit in some ways, and I’d rather check examine, but consensus is a nice stopgap for the great many things examine hasn’t had time to investigate.
For general users, https://phind.com/ and https://neeva.com/ and https://komo.ai/ are some other ai search engines, in approximately descending strength. None of these compare to metaphor in strength, imo.
---
I shared a bunch of stuff like these in https://www.lesswrong.com/posts/ozojWweCsa3o32RLZ/list-of-links-formal-methods-embedded-agency-3d-world-models

the gears to ascenscion 30 Jan 2023 18:51 UTC
2 points
0 ∶ 0
in reply to: ChristianKl’s comment on: 2022 Unofficial LessWrong General Census
they did make a request for comments, which you can see in the backlinks.

Call for submissions: “(In)human Values and Artificial Agency”, ALIFE 2023

the gears to ascenscion30 Jan 2023 17:37 UTC

27 points

3 comments1 min readLW link

(humanvaluesandartificialagency.com)

the gears to ascenscion 30 Jan 2023 15:59 UTC
0 points
0 ∶ 0
on: What I mean by “alignment is in large part about making cognition aimable at all”
it seems to me that a key problem is how to avoid distilling goals too early—in particular, when in a society of beings who do not themselves have clear goal slots, an AI that is seeking to distill itself should help those near it participate in distilling themselves incrementally as well, because we are as much at risk from erroneous self-modification as augmented AIs. the reason to prevent (excess of?) conceptual steganography is to allow coming into sync about intentions. we want to become one consequentialist together eventually, but reaching dense percolation on that requires existing powerful consequentialists to establish trust that all other consequentialists have diffused their goals towards something that can be agreed on? or something? (I think my thinking could use more human train of thought. self unvoted for that reason.)

the gears to ascenscion 29 Jan 2023 23:57 UTC
2 points
0 ∶ 0
in reply to: Trevor1’s comment on: Tools for finding information on the internet
kagi is a paid service. I don’t really understand brave’s model.

the gears to ascenscion 28 Jan 2023 7:36 UTC
7 points
0 ∶ 0
on: the gears to ascenscion’s Shortform
things upvotes conflates:
- agreement (here, separated—sort of)
- respect
- visibility
- should it have been posted in the first place
- should it be removed retroactively
- is it readably argued
- is it checkably precise
- is it honorably argued
- is it honorable to have uttered at all
- is it argued in a truth seeking way overall, combining dimensions
- have its predictions held up
- is it unfair (may be unexpectedly different from others on this list)
(list written by my own thumb, no autocomplete)

these things and their inversions sometimes have multiple components, and many people only allow some of these labels to be applied by close friends, or under specific conditions

the gears to ascenscion 27 Jan 2023 23:11 UTC
7 points
0 ∶ 0
on: WaPo: “Big Tech was moving cautiously on AI. Then came ChatGPT.”

Character.AI founder Noam Shazeer, who helped invent the transformer and other core machine learning architecture, said the flywheel effect of user data has been invaluable. The first time he applied user feedback to Character.AI, which allows anyone to generate chatbots based on short descriptions of real people or imaginary figures, engagement rose by more than 30 percent.

uhoh.

I want to highlight that there is a business incentive directly against safety for the user here. it requires human judgement to simply … not do bad thing too much, with business built on doing bad thing.

the gears to ascenscion 27 Jan 2023 8:04 UTC
−21 points
0 ∶ 0
in reply to: Duncan_Sabien’s comment on: Basics of Rationalist Discourse

the gears to ascenscion 27 Jan 2023 7:58 UTC
1 point
0 ∶ 0
in reply to: Duncan_Sabien’s comment on: Basics of Rationalist Discourse
it says forty three minute read on this link that you want me to use to introduce people to a concept. trivial inconveniences matter for raising water lines

the gears to ascenscion 27 Jan 2023 4:13 UTC
0 points
0 ∶ 0
on: Basics of Rationalist Discourse
for a basics, this post is long, and I have a lot of critique I’d like to write that I’d hope to see edited. However, this post has been posted to a blogging platform, not a wiki platform; it is difficult to propose simplifying refactors for a post. I’ve downvoted for now and I think I’m not the only one downvoting, would be curious to hear reasons for downvotes from others and what would reverse them. would be cool if lesswrong was suddenly a wiki with editing features and fediverse publishing. you mention you want to edit; looking forward to those, hoping to upvote once edited a bit.
unrelated, did you know lesswrong has a “hide author until hovered” feature that for some reason isn’t on by default with explanation? :D

the gears to ascenscion 27 Jan 2023 0:45 UTC
0 points
0 ∶ 0
on: Some Thoughts on AI Art
Video embeds for relevant videos—first the stilted conversation between random frantic ai nerd who is trying to clarify that there’s nothing that can be done to stop ai and we better hurry (I agree with him, he didn’t make it clear enough how hopeless it is to stop it, though, too many people are like “why not just not?” and don’t understand why that’s … nigh on not permitted by physics)
And a couple of related videos I’d recommend, both from the past couple of days. Both are best watched 2x speed with captions imo. Or toss them in whisper and just read the video. It’s good research regardless, these are just blog posts, it’s just that most blog posts are videos because videos get more normie engagement. Sorry.
(And as usual I try to be a hub of “stuff people should have been aware of already”, spider links manually from my userpage or dm me for more links. Basically, holy shit check out IPAM.)

the gears to ascenscion 26 Jan 2023 23:36 UTC
5 points
0 ∶ 0
in reply to: Dagon’s comment on: Some Thoughts on AI Art
I’m not sure what short timeline bettors you’re thinking of here, but I personally think that ai art is pretty much the only form the ai safety problem will ever take. Art is a generative model’s paperclip.

the gears to ascenscion 26 Jan 2023 17:30 UTC
7 points
0 ∶ 0
in reply to: Duncan_Sabien’s comment on: Sapir-Whorf for Rationalists
galaxy brain take that just fluttered through my mind and might be worth ignoring and/or not ignoring: what if every time you coin a new word/phrase, you had to get a new domain name for the word/phrase and make a website succinctly introducing people to the concept. see, eg, https://www.microsolidarity.cc/, for a website that I already use as “obvious-things-documentation website that defines a word clearly and also is a good word according to me”

the gears to ascenscion 24 Jan 2023 23:15 UTC
2 points
0 ∶ 0
on: Some questions about free will compatibilism
it’s a question of where the information is combined, and what options the system considered. yes, the chess engine does have a specific slice of free will about its actions; but clearly, it doesn’t have meta free will about what kind of actions to take. it is relatively quite deterministic in decisionmaking; we have lots of incremental steps of noise on a saddle point.

saddle points are edge of chaos; balancing on them is a delicate act that cells walk at all times by staying alive, so in this sense, life almost always has at least a little bit of free will at the cellular scale. When neurons communicate at scale successfully, they are able to form large networks that represent the shape of the outside world in great detail, and then balance on the decision and make it incrementally, thereby integrating large amounts of information. That means that free will is the chaos resolution process, and is able to integrate large amounts of information while still retaining the self-trajectory.

deterministic or not, it’s highly chaotic—even if the RNG is pseudorandom instead of random, it looks random from where we sit, many levels of nesting larger than it. Because of that, even if it is in some sense pseudorandomly predetermined, it is not known to us, and so we have the hyperreal decisionmaking process of writing some portion of the future; as far as we are concerned, the future is logically underdetermined until we decide what it is via our information integration into prediction and decisionmaking about how to diffuse away some paths through the decisionmaking.

This is a narrow form of the sparse multiuniverse hypothesis: we are bubbles in a potentially mostly-dense multiverse, but within the bubble where we continue to exist, we decide what is most probable by denoising towards it. when someone has a brain injury of those kinds, they can lose some capability to combine information because some paths through the network of brain cells are not maintaining the same various parts of being a hybrid model based reinforcement learner as well. but they retain free will within what they’re able to model.

in the case of the brain tumor case, I would say that the brain tumor causing its output behaviors to defect against other life clearly came from the brain tumor. however, I don’t like the penalty example; my view is that the decision to murder made in clear mind should also be considered an unwellness that may be requested to be self modified away, just like cancers that make one a danger to others. And I would say that in both the case of cancer and of “mere” decision, the question should be one of decisionmaking after the fact: will you allow society to request that you prove you’ve self modified away from the decisionmaking matter-process that caused you to do this last time?

the gears to ascenscion 24 Jan 2023 0:55 UTC
4 points
0 ∶ 0
in reply to: grabbag’s comment on: Brief thoughts on reasonableness of conspiracy
then I don’t think we meaningfully disagree on what there is to protect self and community from—malice incentivizes divide and conquer, so if you’re seeking to protect, unite and aid. my only objections to conspiracy theorism is that it can cause one to mismodel the coordination patterns that cause the behaviors one notices. in general, I don’t think the media as a whole tries to convince people of things; as a whole, mostly they try to get people to click on things. subgroups of media do have specific agendas, but for the most part the shared directions in behavior space do not result in reliable coordination. I also agree with Scott Alexander’s take about how the media usually lies: by omission. I don’t think it’s an awful idea to occasionally read or watch stuff produced by media companies, it’s just important to do citation tracing. don’t be a crackpot; there are always many possible causal hypotheses.

List of links: Formal Methods, Embedded Agency, 3d world models, and some tools

the gears to ascenscion23 Jan 2023 18:21 UTC

9 points

1 comment2 min readLW link

the gears to ascenscion

Call for sub­mis­sions: “(In)hu­man Values and Ar­tifi­cial Agency”, ALIFE 2023

List of links: For­mal Meth­ods, Embed­ded Agency, 3d world mod­els, and some tools

Call for submissions: “(In)human Values and Artificial Agency”, ALIFE 2023

List of links: Formal Methods, Embedded Agency, 3d world models, and some tools