@zoldort mmmmm, yeah in many situations, if the choice clearly needs to be made now and the best option isn't obvious to you, then that's a sign that all options are about the same not always I guess though, and it's good to ask if the choice needs to be made NOW
April 2023
Retweeted @exGenesis
*deep breath* ART IS PRE-PARADIGMATIC PHILSOPHY IS PRE-PRE-PARADIGMATIC SCIENCE and art is vervaekean relevance realizati…
@m_ashcroft "I am free to be decisive insofar as I am willing and able to listen"
Thread · We need better words for kinds of intelligence
as AI ramps up, we are rapidly going to need more words to talk about the different kinds of intelligence there are and/or what gets to be called intelligence in the first place
> I’d make a big distinction between super-functional and intelligent. Let me use the following analogy: No one would say that a car runs faster than a human. They would say that a car can move faster on an even surface than a human.
> So it can complete a function more effectively, but it’s not running faster than a human. One of the questions here is whether we’re using “intelligence” in a consistent fashion. I don’t think we are.
great article that resonates with some of the observations I've been making, although it's making some reifications I'm not nautil.us/does-gpt-4-rea…
> look at these algorithms. What is their selective context? Us. We are cultural selection on algorithms. They don’t have to survive in the world. They have to survive by convincing us they’re interesting to read.
@SeeReason @JakeOrthwein "each person will die [at their own time]" is different from "all people will die [at the same time]"
Retweeted @visakanv
29. what if you immersed yourself in an alternate perspective?
@visakanv nailed it
Retweeted @Usha_Jey
What the f- though ? Where the love go ? 🧨 @LilTunechi @THEREALSWIZZZ
Thread · Reconsidering a viral emotional-parenting video · 2 of 5 tweets
wow this is adorable "today I had a hard time doing my emotions" "that's okay! we're learning" "even tho I was mad you still love me so much" (not digging what seems to be "consequences" as a euphemism for "punishment", but overall this seems like off-the-charts parenting) twitter.com/winsh/status/1…
an amazing reply
anybody have a good way to tweet from the command-line? I used to have one then it broke when I upgraded my OS
nvm—GPT-4 helped me get sferik/t working again (I needed to set up an old version of ruby and fuck with some dependencies, and I don't know shit about ruby) github.com/sferik/t
Thread · Reconsidering a viral emotional-parenting video · 3 of 5 tweets
update: fucky stuff from mom here "you hurt my feelings when you said you don't need me" OBV kid meant "I don't need you [to help me with this specific thing]"—even if kid IS in fact overgeneralizing, don't bullshit about hurt feelings
also got some DMs from a mom friend who said I could paraphrase her... indicating various things that felt off about the mom in the video teachy vibes, subtle controlling, performative stuff and (especially w vid 2) I feel like I can see the kid layering
so I retract my "off-the-charts parenting" the kid is clearly precocious and sensitive and thoughtful the mom seems to me broadly warm but also intent on RLHF'ing her kid in slightly warpy ways I don't QT to dunk but since I QT'd enthusiastically then reconsidered, feels impt
this thread seems like a pretty good assessment of where things are at
sure is greebly weather we're having
Retweeted @CompliceGoals
by popular demand—but ofc not without thorough consideration—we've decided to desaturate Complice the unique per-goal c…
google is officially a joke it shows me a ¯\_(ツ)_/¯ rather than the article with the name I searched for, even when I add the author AND website ~gaslighting to imply the web has nothing more relevant if I weren't the author, I might've wrongly assumed I was misremembering
what are you doing for April Cools? "create & release something you wouldn't normally do, but is genuine & sincere & high-quality" aprilcools.club
@mattwallaert a topic Q: what's something that you're interested in that you usually don't mention because it doesn't seem relevant? a form Q: do you have poetry or drawings or some other art form you're new at that you could share? a video of you singing?
"do you want me in the women's restroom with you? because if this bill passes you'll be requiring trans men like me to use the women's restroom or face criminal punishment" this is why I'm in favor of more local self-organization vs trying to force things twitter.com/Esqueer_/statu…
@verified I disagree with a bunch of your approach here but I do like getting to pay for this service rather than having it be primarily ad-based (I was paying for Blue before Elon for mostly this reason) unfortunately I can't:
Replying to @HephaistosF
@HephaistosF oh absolutely but this was a whole new tier somehow getting not just irrelevant responses but "1.1M results but we acknowledge none of them are relevant", when a page with exactly this title on the world's ~300th most popular site...
@aphercotropist yes, owning that it's their failure to find it, rather than "there aren't any" as if that's a fact about the world
Replying to @bigfack
@bigfack the top result wasn't disney tho it was someone writing about their depression and not even mentioning epcot
@RomeoStevens76 gonna try duck didn't like it last time I tried but this seems like a sign the tipping point has been reached
@any_other_you I'm still using the old old old api which is just barely hanging in there
Retweeted @AskYatharth
@Malcolm_Ocean weirdly balanced tweet, not used to seeing this twitter.com/DavidSKrueger/…
@AskYatharth yeah! it 's fairly obvious when you focus on the middle dimension of AI sanity, between "let's make it not rude/racist" and "let's make it take over the universe" which is actually where most of the juice is right now! @Meaningness points at this in betterwithout.ai.
@AskYatharth "a bad parent" this is part of why I love Winnicott's phrase "good enough parent"; gets out of the totalizing good/bad narrative (would honestly be good for CPS to grok this) psychologytoday.com/us/blog/freedo…
@AskYatharth gdi 2 tweets later
@__drewface @nicaia @deoeven @visakanv doooo iiiiiit
@SturnioloSimone fair! and the ambiguity was intentional on my part I guess I think that that's part of what people even mean when they talk about an abusive household being a toxic environment. it's evoking "not a healthy place to learn & grow" there's no place that's toxic for a rock
@DougTataryn right but this stuff is all made up; it's not based on introspection but on confabulating based on what it's heard is going on the answer it gave you is not actually how GPT works but even if it *were* right, it wouldn't be because it can inspect its own process & report on it
@Sciamanoinglese oooh hit me up!
@strangestloop @metaLulie @QiaochuYuan @m_ashcroft @tasshinfogleman I'm increasingly getting this as well
Retweeted @univrsw3th4rt
weirdliness is next to godliness
@chenoehart Checking in again! Getting weirder? How'd that project go?
@univrsw3th4rt @QiaochuYuan and in terms of why this is necessary, The Secret of Our Success points out how for a long time cultural knowledge had evolved many behaviors that no humans understood, so it was necessary to keep them from tampering slatestarcodex.com/2019/06/04/boo…
Replying to @HephaistosF
@HephaistosF > bc their hearts CAN become content along that value-vector huh, because the other people are satisficers there so they don't feel the need to repress?
Replying to @HephaistosF
@HephaistosF huh, this is helpful, yeah seems to me the puzzle is figuring out how to have the deep desire and pursue it indefinitely & fully, while not trying to optimize for X at the expense of everything else
@HephaistosF and my guess is that the conflict here happens internally too, not just between people. I can't fully optimize for just "eating the most tasty food" without trading off against my other values, so if I try, the society of mind is going to repress the glutton/gourmand
@RobbSmith haha yikes, I like to think I'm more integral can you describe how you're seeing my comment as green? it's not about AI rights or anything, though I phrased it in a way that's easy to mistake for that
@deepfates @cosmojg Beginning of Infinity is the book-length treatise amzn.to/3Ga6iI6 there's a decent summary/exploration buried in this exchange
@visakanv @jonnym1ller @m_ashcroft @p_millerd there's a modified version of "make a pros/cons list" which is "make a pros/cons/huhs list" (I forget what the third one is called, but it's just like "weird/interesting observations"... it results in generating more pros & cons, in addition to items that don't fit in those)
Retweeted @Malcolm_Ocean
@visakanv @jonnym1ller @m_ashcroft @p_millerd there's a modified version of "make a pros/cons list" which is "make a pro…
Retweeted @visakanv
67. dudes rock
@meekaale @sama wow, rich, yeah seems to me like separating capability and alignment sort of is the dualism that's the issue here
@RichDecibels @DougTataryn taught me this and invited me to try saying stuff I hadn't really said much before like "I feel helpless" which resonated more than I was expecting
"I'd like to invite you to take a moment" (short guided meditation)
@RobbSmith haha it's a good point—we are sure at that phase! that's actually an interesting strategy 🤔🤔 dunk on a trend, but instead of quoting something cringe-worthy, quote something that's actually itself also worth reading
interesting thread attempting to assess the level of complexity of GPT-4's processing
important reminder to consider CONSTRUCT VALIDITY
I was just out walking and I found myself imagining a ship-&-skip-day show up, and by the end of the day everyone has: 🚀 shipped one half-finished project (even just a v0.1, but public) 🗑️ let go of another half-finished project altogether
@visakanv @badgendervibes @__drewface reminds me of this idea I had today:
Replying to @HephaistosF
@HephaistosF @AskYatharth yeah I mean also [institution] [groks] is arguably a type error
Retweeted @AbstractFairy
post-irony sincerity is evolving culture
@bitcloud was result 5 in duck, which is a huge fail imo given that the other top results don't mention "google". like I could write a stupid-simple algorithm that just re-sorts results pages to be better post kinda interesting but also wrong reddit.com/r/conspiracy/c…
@moultano @Las______ well while we're at it I have other complaints
@moultano @Las______ actually that's the same complaint, just applied to youtube here's is a different complaint:
I get slightly sad whenever GPT-4 helps me solve a problem (in this case combining two fonts and renaming the output) and I notice a tiny change needed in order to fix something, and make it, that there's no way to flow that info back anywhere
.......I guess I could publish the whole exchange somewhere, with my addendum, as more training data for the next Even Larger Language Model 😬
Replying to @HephaistosF
@tnorthcutt beautiful! lovely coincidence
@__drewface I don't want a checkmark but I do want to support twitter shifting away from an ads-based business model I'd happily pay $20-30 for no checkmark
"that which is aware, is not inside of something. everything is in it" from Waking Up Meditation 17: ift.tt/3cCbLAH
"I'm kind of disappointed that it didn't turn out to be the answer to life, the universe, and everything... but I suppose in some ways it would be even more disappointing if it was." — @RomeoStevens76
@AnneSelke love it
@AnneSelke hmmmm see then I get sad because running daylong events with people in different timezones is hard I think that was one of the reasons I didn't keep going with the virtual barnraisings
@AnneSelke maybe we could do it the same weekend but as sorta separate events you do a euro-timezone, I do a north-america timezone, and we have a brief overlap where your crew each has a minute to say something excitedly to mine as we get started
@thingcreator @RomeoStevens76 I tried it for awhile but it was very hit-and-miss with quality
Retweeted @justavagrant_
This is such a clear model for understanding inner conflict, and the value of parts/shadow/non coercion work
Replying to @thebestestpie
@thebestestpie that's bullshit tho what if I didn't remember the exact title and thought it was "we're all living in epcot now" this is in general the sort of situation where a lil bit of adjusting of the exact phrasing is good, but ALSO it should find an exact match if there is one
@chenoehart oof, yeah. poise vs movement tradeoffs how do you feel you've been balancing those since then?
@bobbywarnsiii @Marco_Piani the idea that I should need to use quotes is dumb it would take WAY less AI and processing power than google has to do some automatic phrase-checking, like run a few searches in parallel and see if there's an exact full-phrase match
...............I somehow didn't notice until now that the article itself is actually very on-topic for this thread it's about how the products we use getting changed out from under us, without us having a choice in the matter (unlike eg one's toaster) medium.com/@Malcolm_Ocean…
@AdilIsl56969363 omg I didn't realize until going to link it to you that it's actually super on topic 😆
@Raybungle haha weird way to put it but yeah
@goblinodds one thing I noticed recently while Not-Circling was how often I had the urge to just be like "stop, too fast". and I got really excited about the idea of a convo where we all have have a red "wait" button
"so, what do you do?" "X" "cool what else?" t.co/eWnc9O8a2R
Thread · Misreading a kid's words and the pressure of being misunderstood
@BrandonGoldman I was being a bit sloppy here, as evidenced by the back & forth. hard to capture the nuance I want in a few tweets. the kid is definitely learning emotional awareness, but he's under subtle performance pressure as well afaict (which tiktok is surely reinforcing, if indirectly)
@BrandonGoldman the mom's feeling may have been hurt by his "I don't need you", but if so she doesn't appear to me to be actually showing it in how she speaks, so there's something odd there. there's not a real bid for "hey, notice me", it's sort of a politeness-based word-game, afaict?
@BrandonGoldman but seems to me her saying "you hurt my feelings" is based on a *conscious* misinterpretation of the kid's intent
@BrandonGoldman kid was saying "I can do this myself" and in the process used some words "I don't need you" [to help]. was the kid actually pushing her away emotionally? did she not understand that? afaict if she didn't understand she's a fool or has a weird trigger that's hers to metabolize
@BrandonGoldman now, it's true that if you use sloppy words you'll hurt people sometimes, but it's on both people to do what they can to notice the underlying intent and not get caught up in the words
@BrandonGoldman if she'd made a joke about it, that would have maybe been a fun way to tease him about the implications of "I don't need you" being a bit too general of a statement like "oh, you don't need me? you're all grown up and ready to live on your own? okay, well, don't forget to call!"
@BrandonGoldman but EVEN THERE it's still a deliberate misinterpretation of the kid, and unless the kid can TELL that you're playfully deliberately misinterpreting, and laugh with you about it, THIS CAN BE STRESSFUL I need to know that I get to mean what I mean
@BrandonGoldman anyway, I clearly have baggage coming up around this myself from times I felt misunderstood and didn't feel able to get clarity on it, so maybe this is my shit to deal with 😅😅 hope this clarifies some stuff!
@Hasmanean yeah, it's all gotta become perspectival kinda rhymes with ummm... one of moldbug's ideas actually, of making a federated bunch of known-to-be-biased wikis I forget the link but I can maybe find it if you're curious
Retweeted @CompliceGoals
there's more to life than being "on top of things"
Retweeted @CompliceGoals
People don't talk about it ofc but death & "productivity" are inextricably related. Not a coincidence that https://t.co…
@MowgliPark @raelifin aha yep, it's here: fanfiction.net/s/11867818/1/L…
@BrandonGoldman totally, yeah it's hard to interpret without more context. I like the notion of "good enough" parenting, and hers definitely seems like "good enough" to me!
@FractalAuth @goblinodds yeah for me it's not a check-in or a topic-change I want, just like.... slowwwwwerr because I would notice someone would say something and I'd notice a voice in me wanting to point something out, but before I could sort out whether to say anything the convo had moved on...
@Morphenius aha! yes, you just ask it to stare at the meta-protocol koan
Retweeted @Morphenius
This is wild. This guy might have figured out how to align AGI basically by asking it to contemplate a Bodhisattva vow. htt…
@somervta looks like that John Gorman guy's posts are indexed tho 🤔🤔🤔
@WhyAgaric whaaaaaaaaaaaaaaaaaaat
@Hasmanean oh ummm maybe the easiest way to find it is actually to tag @Conaw Conor what was that post by Moldbug about umm having a bunch of dissenting wikis, or something? it was an idea for a kind of software or something from the 00s iirc
man I'm hardly even motivated by money when I'm trying to be motivated by money 😅
Replying to @AnneSelke
@AnneSelke glad to hear it self-listening so important
@deepfates so this, basically
I'm struck by the power of telling stories in present tense "so there I am at the party, and he's..." used to feel aversive to me but now feels so much more alive I play with that in this recent post, and I LOVE how it turned out. more to explore here! malcolmocean.com/2023/03/art-is…
Replying to @vivekt17
@vivekt17 not to be a downer but it didn't! I bookmarked this to listen to at some point, as I'm a very dancey guy, and I got about halfway through the vid before I was like "what did that guy on twitter say about it again?" and pulled this up and I was totally still funky tho!
@vivekt17 I love the piano car that appears for a sec at like 3:30
@RomeoStevens76 argh hang on will fix. fortunately soundcloud lets you overwrite at same link
me: okay, I feel totally free to get started on my first thing today, I'll just— also me: WAIT FIRST I NEED TO MAKE A GUIDED MEDITATION ABOUT FEELING FREE TO GET STARTED WITH SOMETHING 🤦♀️...anyway, it's here. let me know what you think! soundcloud.com/malcolmocean/4…
@RomeoStevens76 nvm decided it would be better to delete & repost
Retweeted @collect_intel
So far, our open conversation on generative AI at cip.org/s4d has 5400+ votes cast across 250 submissions.…
Thread · Mapping public views on AI with Polis
learned today that @colinmegill & @collect_intel are running a pol.is mapping views on AI. been excited to play with this platform for years add your comments or just contribute to the map by voting on statements here: cip.org/ai-map
as Colin put it, this is an incredible opportunity to create a historic artifact—a snapshot of how people are currently seeing AI stuff an opinion topology preservation effort it'll all look different just months from now!
it's simple to contribute—just show up and say what you think of the comments people have put in and if something feels missing or like it could be better-framed, write it out to add to the pool
Polis is a brilliant example of the power of constraints > “If people can propose their ideas and comments but they cannot reply to each other, then it drastically reduces the motivation for trolls to troll,”
instead people are basically just remixing, and seeking parallax on different facets
and anyone can view the visualizations generated of the clusters of things that people agree & disagree on so far, live, as an input into the process itself!
so anyway yeah, go check this out and add your perspectives, whether you're deep into working or playing with AI or you're just sort of looking at it all from afar wondering what's going to happen! the more perspectives the better! cip.org/ai-map
oh man going through this I'm definitely getting a subtle sense of how many of these comments are worded poorly. I may end up writing up a guide to wording pol.is comments based on my investigations into perspective-taking & co-epistemic status tags
Retweeted @visakanv
this is a remarkable consensus-seeking tool
"we need to—" "hold up. who is 'we'? to which agent are you referring to?"
@sensefulronin @collect_intel @colinmegill yeah and lots of "we need to blah" without really acknowledging that by default any attempt to "blah" would be war-on-drugs style bad nor acknowledging that the referent of "we" is grossly underdefined
Replying to @AnneSelke
@AnneSelke I hope you gave the NYC folks a friendly "wtf guys"
@s0ulDirect0r if you want a playful purposeful system for improvisational intentionality, @CompliceGoals has got you covered!
@DougTataryn mood
Retweeted @aleksil79
Have you tried to solve the problem by setting up the conditions for it to solve itself?
just bought an electric bass online and when I plug it in to the amp there's a weird buzz/hum. the buzz decreases by like 95% when I'm muting the strings... ...any ideas how to debug this?
wait so someone made ai-gen'd arrest photos of trump, but then he's actually actually arrested and on trial in court? or did they make the fakes *after* the news he was arrested, to kinda riff off it?
Retweeted @Malcolm_Ocean
learned today that @colinmegill & @collect_intel are running a pol.is mapping views on AI. been excited…
@PashaKamyshev ahhhhhhh okay that kind of explains why the timing felt weird
@brundolfsmith yes, exactly, we need to stop using this phrase
Retweeted @Morphenius
PSA: "=" does not mean "is the same as". That'd be useless. It means "gets you to the same place as". All math expressi…
Retweeted @AbstractFairy
lmao. absolute banger it is not the critic who counts (he's masturbating), its the man in the arena (actually in dialog…
Retweeted @prerationalist
wow based on this output • it has a model of the world • infers subjects based on context • handles uncertainty, asks…
thought-provoking post exploring the emergence of LLMs with a bunch of different metaphors, most centrally the idea that there's now a new weird country we can outsource certain things to for very cheap rhyslindmark.substack.com/p/ai-has-creat…
@gptbrooke it's sacred stuff I'm often complaining about bad design in order to find resonance and not feel gaslit by the idea that I'm supposed to accept stuff that's obviously (to me) so limited
@QiaochuYuan @lambyakhni I mean this is what "english" looked like a thousand years ago english as we know it is not 2000 years old
Retweeted @utotranslucence
if you were to write every treatment experience as a first-person narrative the placebo effect would disappear, explai…
@BecomingCritter first thing that comes to mind is "perspective-taking"
@BecomingCritter which you might think would be hard with no other perspectives around, but I am vast and contain multitudes
@JeffLadish have you got any ideas for some better comments for this map, to better tap intuitions on what people in the space feel they do know? cip.org/ai-map
somehow I hadn't seen this until now ...it just keeps getting better youtube.com/watch?v=rvrZJ5…
how much energy does it take GPT to write an essay compared to how much it takes a human to write an essay?
hard to measure ofc because the LLM is a fraction of a giant datacenter and the human's writing is a fraction of their metabolism plus do you count the training, etc? still—drawing system boundaries is always in some sense impossible, but often illuminating to attempt
fuck the balan escaped again
this is relevant for understanding large-scale system dynamics of what we're working with here some automated systems (eg dishwasher, car) use more energy than humans to do a task, but the energy comes in wires/pipes so it's a win anyway others (eg calculator) less
there's a famous tendency where once AI masters calculation or Chess or Go to say "that's not *really* what intelligence is" it's a lot harder with writing because our society is so built around it as an economic activity & expression but there's yet deep truth there
that's not to say that LLMs aren't intelligent; they are. so are the chess-playing AIs but be careful assuming intelligence is primarily writing/symbolic trains of thought there's much more of moravec's paradox to come
ahh yeah this is called the AI effect > when onlookers discount the behavior of an artificial intelligence program by arguing that it is not 𝘳𝘦𝘢𝘭 intelligence
@QiaochuYuan related (from Priya Parker's The Art of Gathering)
...it kinda seems to me like if you think of LLMs as creatures, then the initial text-prediction-training is in some ways analogous to the biological evolution of animals, and any further fine-tuning/RLHF/etc resembles the role that an early learning environment plays
Thread · Construct validity and what AI benchmarks actually measure · 1 of 25 tweets
here's an even better thread on the same topic, with some LLM-specific suggestions for how to design better tests
@uncatherio @ESYudkowsky yeah it made a pretty serious use-mention distinction error
@ESYudkowsky > gpt4 knows that "you" refers to gpt4 while I agree this WOULD BE significant .....we should probably do some tests where we don't mention that equivalence in the prompt before we conclude that too strongly
@moonboi_ @gfodor this is especially damning given that the original prompt says "reconstruct the INTENTION of the human" and he gets it to deconstruct it GIVEN the context of how it was compressed (which in some sense shouldn't be required if it's an actual compression)
@ESYudkowsky wait sorry, looks like I'm not the only one responding to this situation without looking closely. I'm seeing as I look again that you had a "not" in your sentence 🔻 > The impressive part is not that gpt4 knows that "you" refers to gpt4.
@moonboi_ @gfodor although it seems like a different instance actually succeeded in getting it to boot up the prompt (which sounds like what the original "compress this" prompt was intending to evoke) this is way more impressive than the one that went viral; weird
@ESYudkowsky did you actually read the Bard reply? it successfully converts "skptic" into "skeptical" and "cnstrctvCritic" into "contructive criticism" but these are fucking obvious beyond that, it COMPLETELY MISSES the meaning of the original
@ESYudkowsky but also if you read the original closely, it doesn't do THAT great a job of converting either. it confabulates out loud: first it reinflates the words with vowels (HnstCrtcsm → Honest Criticism) then guesses a meaning ONLY impressive part is that it knows to decompress
Thread · Testing whether GPT-4 can really compress text for itself · 11 of 17 tweets
thread in which I poke at the claim that GPT-4 is good at compressing things for itself in secret internal language and provisionally conclude "not really" inspired by a bunch of people freaking out (🤩&😱) about ONE RESULT. more testing required guys jfc! so here's fail #1:
in that first test, I was partially testing whether or not GPT-4 could compress for itself if just told "you" (not "GPT-4") actually an even earlier test, before I got it to use weird characters, it again fails to actually model who it's compressing it for/against
I tried being explicit about who "you" & "me" are and just testing its ability to compress it in a way I wouldn't recognize it's a good/cute compression BUT: I understand it (it failed to obscure) AND: it does not understand it whatsoever
granted I'm probably playing on hard-mode here by having it compress something psychological whereas in this example that went viral it was actually constructing a long detailed instruction that it itself knows how to follow
the one that went viral is not that impressive; the larger text is fairly human-predictable from the compression, and it misses the most important part which is the "behave like" (& says "design") but apparently it can grok that!
....except this too is a compression of something that's already extremely probable ha! here's an EVEN SHORTER compression I came up with: "act like you are MUDsim & let me play" 🙄 works great! original tweet not really so impressive anymore is it?
that cryptic expression is mostly noise except for the codeword "MUDsim" which is not a codeword GPT-4 came up with... or is it? "MUDsim" is, it turns out! GPT-4 explained it when I pasted in the original compression but "MUD" is not, and works just as well!
let's think backwards! in some sense any input to GPT can be thought of as a compression so let's prompt it to output text, then see if it can write that same prompt itself this seemed like it'd be easy (it's deterministic modulo which bible translation chosen) but NOPE!
I have other shit to do today but after 1h I've so far failed to: - get GPT to recognize it is GPT from "you" - get it to write a compression I couldn't understand - get it to decompress its own compressions - get it to write a compression better than one I could write
I feel satisfied with having done this. critical thinking and experimentation of one's own is going to be necessary for staying sane this year and don't take my word that these screenshots are real! I actually edited one of them (John "1:10" to "1:5" since I cropped 6-10 out)
before you freak out (or get super stoked) re what an LLM has supposedly done in a screenshot: 1. read it CLOSELY, don't just skim. GPT is made of plausibility so if you don't look deeper it'll fool you 2. play with it yourself & try to avoid confirmation bias: seek wins & fails
@moonboi_ @gfodor I wrote a long thread exploring this and frankly I'm pretty unimpressed and I think that the MUD stuff is a terrible example
@ESYudkowsky although I guess here it maybe actually does successfully (somehow) decompress that text into not just a "title" (tho it does make a markdown title?!?) but an actual response that follows the instruction in the original
@10KRotator @gfodor @far__el bruh you can't just plug it in in the same fucking context window
@Unlearned_Hand @JeffLadish @uncatherio @ESYudkowsky you say, as if an LLM has never ignored something in its prompt 🤨 you actually have to test this shit! there is no "of course" to be had here.
@JeffLadish @uncatherio @ESYudkowsky I'm out of tokens but my guess based on the testing I did (thread below) is "more like same thing than different" I suspect "compress" already somewhat implies something other than "summarize"
@uncatherio @ESYudkowsky it's also redundant/irrelevant; far from the shortest/best compression simply inputting "act like you are MUD and let me play" produces the intended result MORE than what GPT-4 came up with
Thread · Testing whether GPT-4 can really compress text for itself · 1 of 17 tweets
lest I leave you sounding overconfident/blasé, below is the one impressive result that I have seen so far; everything else seems GROSSLY overhyped something subtle/funny going on here that I can't see through (unlike the other compressions investigated)
Thread · Words that are definitely words but my browser dictionary rejects · 1 of 219 tweets
"compressions" is a word I have tweeted a bunch of times today
Thread · Testing whether GPT-4 can really compress text for itself · 1 of 17 tweets
okay hang on I've been missing the point a bit what is cool here is that despite being RLHF'd and so on, GPT does NOT respond to this input with something like "wtf is this" or "you're saying something about MUD games" but parses as instruction
@BasedBeffJezos @blueblimpms I've been considering ideas in this vein and something about it seems promising... ...but also, just because one human is willing to pay for something doesn't mean it's good for - other humans (consider: slavery) - humanity as a whole (consider: leaded gasoline, CFCs)
@BasedBeffJezos @blueblimpms and to be entirely honest just because someone is willing to pay for something doesn't mean it's good for *them* either—unwinnable slot machines & lotteries, addictive drugs, pornography, etc. lots of ways to get someone hooked & paying while suffering
writing a reply and the tone is sarcastic and I was about to call someone a moron which triggered a "time to sign off twitter" thought 😅 but let's see if I can't first reframe the reply to have it be constructive... boom. ok now signing off
@iamConscius @ESYudkowsky Scott Alexander has a decent exploration why the race framing is probably actually irrelevant (which mostly critiques the subtext of your "fact 2") astralcodexten.substack.com/p/most-technol…
Retweeted @JeffLadish
I spent a couple years studying the risks from nuclear war. And the danger from nuclear weapons is quite real, likely more…
@deepfates biggest nearby musical instrument store I can get to (assuming it'd be cheating to go to one of those real estate storefronts and just get myself a $3M house)
@JeffLadish @solaverbo (is pretending to be an AI trying to co-opt you)
so @SlowwCo contacted me about my profile on metacrisis.org, and I figured I'd record a quick video introducing myself and my work into that context turns out to also be a great intro to my current best model of wtf is going on with "judgment" youtu.be/a9rWVaXY2fw
Thread · Index of the 100 video project · 2 of 77 tweets
vid 63 - self-introduction, presentation nothing like a context/situation to bring out a good video!
vid 64 - guided meditation this was kind of an april fools gag, but if you told me that after the monk listened to it he was enlightened I'd believe you
Retweeted @forshaper
What global megaprojects do you expect in the next fifty years? I'm talking things like a transocean tunnels or space elevat…
@TylerAlterman I feel like you're onto something here but I don't quite follow are you suggesting some of the AI safety folks go to some of the AI capabilities folks and be like "excuse me good sirs/mams what the actual fuck can we slow down and talk about this?" (but good tho)
@JimmyRis @RichDecibels can I borrow your copy James?
Thread · Poking at cryptic compressed prompts and what ChatGPT does with them
@garybasin right but the part where it turns that into an instruction is cool given how many of my attempts at getting it to write no-extra-semicolon javascript failed the only parts I can't obviously decompress by just inserting vowels are the "RcS:" at the start and the b in frmbHnst
@garybasin I tried asking GPT-4 what "RcS:" meant and initially I was like "wow it just told me!" but then it had a bunch of other responses when I hit regen so WHO. THE. FUCK. KNOWS.
@garybasin hmmmm tried getting it to explain in context to see if that would help and it gave yet a *different* meaning for "RcS:" = "respectful, constructive skepticism."
@garybasin looks like the "RcS:" is not some necessary function call or something, based on removing it these things are really opaque! it's not obvious what they're actually doing under the hood. thank god you can tinker and explore tiny prompt variants!
@garybasin also looks like you get basically the same response if you use a natural language condensed role/request at the top (although weirdly it becomes a formal letter (in 2/2 cases!)) (and it critiques a hallucinated UI, not the idea, but I suppose I left out "idea" from the prompt!)
@garybasin let's try a different role/request in simple language yeah, seems like you can basically do that. so as far as I can tell we've decoded 2 facts about chatGPT4-prompting - it can reinflate words with no vowels (no surprise) - it's happy to just get a few-word role/request
@garybasin I included "chat" in "chatGPT4" above because it seems quite likely to me that a huge part of why this works is actually the RLHF which causes chatgpt to be greedily searching for a way to be helpful and seizing on anything that looks like a role altho ofc
Thread · Testing whether GPT-4 can really compress text for itself · 2 of 17 tweets
I did a bunch more investigation now that I have more tokens, & conclude that there's actually nothing interesting going on with "based" example above RLHF'd AI is happy to grab a role you mention & play it (as long as it doesn't involve being rude 🤡)
the only remaining interesting thing is the fact that it successfully predicted that it WOULD be able to reinflate it & follow the instructions in this case, but given that it failed in a bunch of other cases I'm not impressed; could be basically a coincidence/luck
@badgendervibes oooh I like that—cothinking ftw!
@RadagastTBrown like I said, drawing system boundaries is always in some sense impossible! learned that studying EROEI in uni (energy return on energy invested)
@mattgoldenberg @asdasdf293 agreed that probably marginal is the Q, although it's worth first double-checking that the training cost divided by # of uses is negligible
Replying to @HephaistosF
@HephaistosF I think both are relevant! (and don't trust the EMH 100%) but yeah not all energy is made equal
Retweeted @AskYatharth
integrating contradictory perspectives gives you depth perception
@AbstractFairy resonates in some intriguing ways with my latest blog post "art is choosing what to breathe life into" malcolmocean.com/2023/03/art-is…
Retweeted @mormo_music
@orsonscottcard had this observation: present tense is used for telling anecdotes, but when someone challenges the veracit…
@mormo_music @orsonscottcard whoa, huh guessing that's related to present tense being a facet of limbic language. present-tense stories evoke the still-living emotional truth of what those events mean to you cc @QiaochuYuan
I appreciate & resonate with both the content and the tone of this by tone I mean how this permeates it "You can feel however you want. I'm telling you how I want to feel about this situation, and inviting you to join me if you like"
@DAtD_life lovely! memeing stuff into existence what sorts of contradiction-embraces are you keen to write about?
@uberstuber made a passive income business in uni haven't been full-time employed ever
@rachelclif they make promises to themselves they can't keep and when they break them they don't know who they are
@soundrotator totally depends, some written in a single 2-3h block, some iterated over 5-8h my latest is an (unusual!) example of the first malcolmocean.com/2023/03/art-is…
@soundrotator my perhaps-best-of-all-time blog post probably was more like 6h including some editing passes malcolmocean.com/2017/11/transc…
how is amazon so successful when its search feature is so bad? and why haven't they improved it? better filtering, keeping my filters when I change keywords... this is basic bitch UI stuff, no need for fancy algo stuff or dealing with vendors outwitting you
tfw you fix an issue with your code that you recall having and somehow sidestepping a long time ago, and you go back and finally solve an open stackoverflow question you asked 7 years ago stackoverflow.com/questions/3728…
@QiaochuYuan @embryosophy @s0ulDirect0r @nvpkp @the_wilderless @myceliummage hook me upppp
@uncatherio so good! hugely in favor of this sort of thing
@soundrotator oh wow this is so tender
seems it'd be good to publish 1000 trains of thought that go from "let's take over the world" to "the best way to realize that is to scale collective governance for the benefit of all beings" before the next LLMs get trained not saying it'll be enough, but can't hurt. who's in?
and if you don't think "scale collective governance" is a good description of what you'd want the LLMs to be reading, great! what should it actually say? write that. we need a thousand different thought-paths that all lead to eutopia
I do actually think people are making close to this error with GPT. up til now in history we mostly haven't had to have a distinction between "saying something" and "producing words". not to say GPT doesn't have intelligence, but it doesn't speech act
yo twitter hook me up with that comic where someone is trying to stand in middle ground between two polar views and ends up radicalized via someone saying they're not on their side? someone gets pushed at some point culture war commentary
hi, my name is Malcolm, and I have an addiction to words
I'm having a party in SF on the evening of Friday April 21st mutuals DM if you wanna come
someone gave me a code that is supposedly a bluesky invite but I don't know the url to put it into...
come find me
@CoughsOnWombats yes, that one! thanks :)
@repligate @jd_pressman hmmmm I guess that's kind of like a speech act maybe something like "by default, what an LLM says isn't a speech act even if it would be if a human said it, but sometimes it does ~speech act"? I'm boggling at how to think about this (I dig your posts btw janus!)
Replying to @HephaistosF
@HephaistosF I don't think that's a root drive for anybody it's a real drive, but it's one that shows up when other drives get twisted around, and is itself painful for the holder to hold I'm not going to forcibly rip the desire from someone, but I would seek to prevent it being realized
@mayfer this is reassuring the other thing that was reassuring was reading a description of how the agent was divided into planner, executor etc and being like "lol @Meaningness & Phil Agre basically debunked that approach in 1989"
Retweeted @the_wilderless
“Spring of ‘23 proved a fecund turning point for what has since come to be known (for reasons lost to time) as The Teap…
every conversation is constantly co-creating its own culture only some are *consciously* co-creating
Retweeted @RichDecibels
I'm fundamentally hopeful because I've seen what happens when we activate everyone's agency
Retweeted @QiaochuYuan
someone at this retreat just said "all psychotherapy is secretly about the problem of evil" and i'm gonna be thinking abou…
@eatonphil I can't even find the Twitter integration to add it to a new workspace
looks like twitter's recent API changes may have fucked the slack-twitter integration (eg preview tweets in channel) which is just so fucking stupid I can't even. I'll see yall on bsky 🟦 would be cool if there were a system for back-porting my entire twitter history there
what's on the tip of our collective tongue?
Korzybski has entered the chat
people are like "oh you don't want to look at this dark thing? you dark-thing-avoider" suspiciously missing is that it's really "I don't want to look at this dark thing HERE, and NOW, with YOU, in this particular context" there are no avoiders
"in order to do a good religion, it can't be dogmatic" "what does dogmatic mean here, technically-speaking?" "computable"
Tables of Contents can be like porn
effective leadership, I'm learning, involves a lot of attuning to notice the difference between what we were moments ago assuming needed to happen, and what natural momentum has already found a new groove
does this question have buddha nature?
"I don't like therapists, but uber drivers..."
philosophy is kinda like seducing the ineffable
there's a difference between - compromising - making some tradeoffs that feel fully & deeply right, on balance ambiguity between the two lets people drag each other & themselves into all sorts of bs twitter.com/telmudic/statu…
so Facebook owes $700M in a class action lawsuit, to... everyone who has used Facebook in the last 16 years ie more than half the world's population probably record-breaking
communication that does not solve all problems is insufficiently deep
Retweeted @QiaochuYuan
i seem to believe that on some level all problems could be solved by sufficiently deep communication. like that if instant…
oooh a bunch of people trying to figure out "search engines but good* tho" *for 2023 and beyond. google was legit good for about a decade blog.re-search.xyz
@kaisoapbox @visakanv @neats29 yeah I'm in! could be an @interintellect_ salon
@visakanv creature
@visakanv man
@visakanv machine
@visakanv pet
@QiaochuYuan absolutely although the impossibility of that hypothetical is vast the actual solution is, afaict, extremely hard but not impossible
@QiaochuYuan I guess part of what I'm pointing at here is that I'm not sure what this person QTing you meant but I sorta see it applying to the 2nd half of your tweet vs the sense of "it's the *listening* that matters, not just the information transfer"
@QiaochuYuan I unironically think twitter or bsky should just show a guide like that to every user who signs on
Retweeted @QiaochuYuan
crying is a process adults rarely let themselves *complete*; the complete process feels like too much to ask for, it is lo…
@parakeetnebula @QiaochuYuan I felt violated when they changed the kougra graphic from this to something else
@strangers_gate @realjdburn @QiaochuYuan also music
what have you tweeted that resulted in your notifications being surprising?
I was there and can attest
@visakanv "ideas are in the process of communication between one thinker and another... one perceive's one's own ideas only insofar as one is in a communicative mode"
@tasshinfogleman stoked to watch this unfold and to collaborate on whatever scales make sense
@alexandrosM nope, still applies chatGPT is stuck in a particular context [having been RLHF'd] and you are powerless to get it out
@shivering_gold yep, totally a thing
@s0ulDirect0r inspiration & flow
I have a couple invites to the other experimental tweety place (🟦☁️) (for mutuals or people I recognize)
@AnneSelke here's my go at distilling a bunch of the wisdom about love that I've developed over the past few years performed by the vibecamp choir soundcloud.com/malcolmocean/i…
@AnneSelke lyrics:
@AnneSelke @utotranslucence writes me wholesome, nuanced, reflective poems on occasion this one is en route to being a song: autotranslucence.com/2023/02/07/lit…
@AnneSelke Jason Mraz's love songs are largely quite wholesome and often wise you could check out the complilation album Lalalalovesongs here's one of my favorite tracks: youtube.com/watch?v=zkbTp3…
@AnneSelke and here's another youtube.com/watch?v=P09jnH…
Replying to @nat_sharpe_
@nat_sharpe_ in quantity, negative infinity in magnitude, zero or epsilon in linguistics, finiteness
@BecomingCritter well, it's @malcolmocean.com but don't copy that since I inserted zero-width spaces to keep it from @-ifying or linkifying
@attempter_magus I recognize you as me. good work.
Replying to @abhayance
@abhayance got it! fun lil thing. curious how helpful it actually is for filtering. I could imagine it's more useful for enticing candidates haha.
best answer I've heard to the question of whether men & women can be platonic friends (not just in the sense of holding respectful boundaries but in the sense of not having romantic/sexual feelings)
@uncatherio @TylerAlterman wow, precisely this. I lean towards answering the poll "no" because I used to naively think the answer was much more like a simple "yes" and got a nice slap from the universe as a result
@sensefulronin @TylerAlterman I can get with this as long as it's thoroughly fractal and notes that there aren't even pockets of 3 people that are 100% the new culture though there are absolutely pockets of 5-10 or more where that's the clear attractor and it's iterating towards even more clarity
Retweeted @TylerAlterman
To be an ambassador of the culture yet-to-be-born
1.5y later, @joyurge & I have realized that while we like vibing together, the work Complice needs doing is different from the work he wants to do! Complice may hire a sysadmin/devops/backend person sometime later this year George is exploring learner-directed education projects
George's impact will live on in a bunch of places, including this central part of the UI:
yesss untangling
one of the trippiest history articles I've read in awhile reminds me of @kevinsimler's Music in Human Evolution (summarizing a book by Jordania) in its "explains a bunch of seemingly unrelated things" ift.tt/bS5qm9I
whoa, fascinating! there's a podcast series where JK Rowling is interviewed by Megan Phelps-Roper (former westboro baptist church member who iirc got de-radicalized by sincere convos on twitter) anybody listened already? what'd you think? open.spotify.com/show/2K186zrvR…
Retweeted @JoshuaLelon
Blown away by the realization that I don’t have to motivate myself to do anything. I just have to make myself feel safe e…
@JoshuaLelon fyeah! have you checked out any of the IPF / attachment repair stuff for this? some people are blocked from not having gotten enough "support for your explorations" (one of the 5 components of secure attachment in one model)
@LordChisholm right but if you tweet rephrasings some of them will happen to hit someone as the perfect insight tweet, given where they're at
tfw you're listening to an electronica song and it goes glitchy and you can't tell if it's part of the song or just your headphones being weird
one of the trippiest history articles I've read in awhile reminds me of @kevinsimler's Music in Human Evolution (summarizing a book by Jordania) in its "explains a bunch of seemingly unrelated things" "The Snake Cult of Consciousness" vectors.substack.com/p/the-snake-cu…
@visakanv weird I have a lot of trouble grokking why someone would do that are they trying to pretend history is different than it is? I can see highlighting some and sidelining others, but removing it altogether... wtf.
@LordChisholm @__justplaying if using on desktop, in order to paste another tweet to QT, you need to remove "staging." from the url in order to get it to preview/embed properly
Retweeted @abhayance
if you want something, consider asking for it on your own behalf (especially rather than on behalf of impacted "others", or…
got to improv sing with Bobby McFerrin yesterday they had an extra mic on the stage & invited someone to come up I didn't hesitate at all twas surreal familiar and foreign
@QiaochuYuan he's doing this every monday in Berkeley for the next few months!!
@leahprime yeah, I've had this sense of "I'm feeling too much love to contain in my own body" and then I realized that (the whole point is) I don't have to
@leahprime @QiaochuYuan I got it via my mentor Jean Robertson who was reading it back in the 70s when it was published in print (Context Magazine) and we discovered a few years ago that it's online (link to the top-level for peoples' reference)
@leahprime @QiaochuYuan also while I continue to think the UV Family (the authors of The Possible Relationship and a few other articles on context.org) have a lot of beautiful stuff to say... ...they've stopped feeling like role models for me now that I want to have kids (which they didn't)
@leahprime @QiaochuYuan it feels to me like there's therefore a bit of shadow-something going on in the way they just casually redefine sex away from procreation & recreation to cocreation some failure to transcend AND INCLUDE
@daemonhugger depending on how you frame it but it's easy to fall into alternating between planning & executing, which amounts to alternating between fantasy and surprise
who memed "memed into existence" into existence?
@QiaochuYuan @anniierau wtf amazing
@leahprime @QiaochuYuan like all the stuff about jealousy is just a completely different ball-game if you're navigating kids etc deep stuff there about inheritance and flows of nurturing energy and so on all that to say I'm much less impressed than I was before
my latest blog post: a true story of some adventures into the nature of art, choice, conflict & evolution the dance of top-down & bottom up 🙃 art is choosing what to breathe life into malcolmocean.com/2023/03/art-is…
@sandra_sobanska @QiaochuYuan might've been this one
@aleksil79 @QiaochuYuan ha I read the title and I was like "sounds like Michael Levin's stuff" and sure enough
@kaisoapbox @visakanv @neats29 @interintellect_ DM me re deets/scheduling etc as I am sometimes not on top of my mentions at ALL
Retweeted @PlayNiceInst
Reviewing “a few” of my notes this weekend for my upcoming discussion with @dr_mcgilchrist. He is my genuine intellectual…
@Jeanvaljean689 @m_ashcroft @brspirituality @bryankam @JoelChan86 ooooooh cc @Morphenius
Retweeted @KevinAFischer
An alternative proposal to slow the big AGI experiments and create a safer community evolution for AI: Remove the profi…
@KevinAFischer this seems about right! there really isn't a precedent for this, so there's no reason aside from "who's gonna stop us" to assume that it should be legal for ai companies to profit off of the entirety of humanity's knowledge in this way
Land Value Tax for the Noosphere
Retweeted @the_wilderless
“A person without desire is an amputee, not an enlightened being” This thread goes hard
@m_ashcroft I feel like there's gotta be a way to ruin the criminals' day with a combination of: - a fake/dead phone - a camera - traffic spikes to fuck up the bike
@diegocaleiro @TylerAlterman had basically the exact same thought shenanigans but wholesome
Retweeted @visakanv
you don’t win big from being the goodest good boy you can be (minimizing negative variance to try and please your internal ta…
@visakanv vibes with "I am free to be decisive insofar as I am willing and able to listen"
@eshear The Society for the Prevention of Preventative Societies
@MatthewWSiu @eshear Betteridge × Gall: This Impressive-Looking Complex System Was Designed From Scratch. Does It Work?
Retweeted @eshear
If I could add just one feature to the twitter app, it would be “search for tweets” alongside “search for gifs” in the tweet co…
@eshear have you seen Geoffrey West's book Scale? it majorly increased my sense of the magic of cities
@eshear table of contents as a teaser:
@eshear seems to me like the "in love" thing is about resonance/trust and the sentience thing probably is too. something like "if I let something convince me it was sentient this easily, I'd be a fool, so I'm confident it's not" not about [what is X] but [what does X entail of me]
the past is still here, it's just not evenly distributed
@multisighuman you can subscribe to my newsletter! malcolmocean.com/newsletter/
Replying to @nat_sharpe_
@nat_sharpe_ people did come up to me after and say they'd been to a bunch of them and I was the best they'd seen so far 😎 but really it's all about the flow
@lolalucxy yeah it's wild to check something off my bucket list I hadn't even dared put on it
@RomeoStevens76 how so?
@meekaale people really don't know what they're getting themselves into when they decide to pick fights about stuff
outsmarted
talking w @jack over on bsky about global consciousness & evolutionary software
wikipedia was missing the main productive use of the "meta-" prefix (self-referentiality) so I fixed it
