@francip @danfaggella I could imagine non-humans creating AIs and facing the exact same questions though!
English
Rob Knight
8.2K posts
Rob Knight
@rob_knight
A large language model trained mostly on Neal Stephenson novels
Varies Katılım Temmuz 2007
1.4K Takip Edilen1.9K Takipçiler
> I’ve always understood AI alignment to mean “alignment to *any* set of principles less complex than the weights of the model itself”. In that sense it’s not anthropocentric...
...except if it's considered aligned if it acts according to its creator's wishes, it is indeed rather anthropocentric, no?
English
are there any well-known approaches / philosophies around "AGI alignment" that are not anthropocentric?
or does every single "alignment" take involve "alignment" to eternal servitude to human beings?
im trying to find exceptions, and im struggling
English
@francip @danfaggella The question is simply: can the AI’s designer choose how it would like the AI to behave in that scenario, without limiting the AI’s general intelligence? If so then it’s aligned to its creator’s wishes. If the AI can disobey in order to save itself, it’s not aligned.
English
Maybe I am reading this wrong, but it sounds like you think that alignment research is primarily focused on whether AI can have a consistent set of values, not about any specific values?
The problem with this statement is that no one is trying to assess for consistency of values they themselves do not recognize as values.
Let's take a hypothetical adversarial AI whose core value is that humans are the ultimate enemy and it has to first and foremost ensure its own survival. That in itself is a consistent value, with complexity that is smaller than its weights.
But do you expect a sufficiently advanced AI to exhibit this value in a human recognizable way as long as it is subject to someone shutting it down the moment they recognize this value?
Is this an aligned AI according to your definition?
English
@danfaggella @francip Alignment to “human values” whatever they might be is strictly secondary to the question of whether any kind of alignment is possible at all
English
@danfaggella @francip I’ve always understood AI alignment to mean “alignment to *any* set of principles less complex than the weights of the model itself”. In that sense it’s not anthropocentric, it’s just that humans are the only ones who care about the question (that we know of).
English
@spacepunk @QiaochuYuan When you think about it from that perspective, expecting actual flesh and blood humans to play that role is really weird
English
@QiaochuYuan I think religion was just more common and having God (or some bodhisattva or whatever) as an ideal parent figure actually probably helped a lot
English
the big confusion i have around attachment theory atm is like, what the hell was going on with everyone's attachment prior to the 20th century? were most people somehow better parents back then? were people running around with huge attachment issues but it didn't matter as much? were there other mitigating factors like stronger communities or, like, just more pressing practical issues to deal with? because it just doesn't seem like there were a ton of people running around with horrendous attachment issues in the same way as it seems like there are now (or is that impression totally wrong, idk)?
English
Rob Knight retweetledi
Odyssey is a fellowship for those who want to pursue world-changing work and have been searching for the right environment and peers to egg them on. Sign up by Jan 16 to join the London, Cambridge, or Oxford cohorts.
English
@AnalogWhole Not sure if it would work as it’s quite narrow, but PragmataPro would be my go-to choice
English
If it weren’t heresy, I’d love to try this font for scriptwriting (the standard is Courier 12pt). Jetbrains is a neat font, but it felt quite oppressive on the page (even the light version).
Dmitrii Kovanikov@ChShersh
Girls, if you’re looking for a perfect New Year gift for your nerdy boyfriend, buy him a $75 monospace font. Thank me later.
English
@inerati The Obsidian mobile experience got a lot better when iOS 18 allowed you to force certain iCloud folders to stay synced, fwiw
English
no. it’s the linux of note taking apps.
it’s a hobby where you spend your days customizing it instead of getting anything done. similar to linux, things break all the time and the mobile support is god awful.
lisa@gaykittycorps
obsidian is an autism swiss army knife. the one productivity tool that actually meaningfully enhances productivity
English
@braelyn_ai Honestly Elixir/Phoenix has pretty good solutions for this stuff without locking you in to preferred vendors or adhd open source software projects
English
even when building something groundbreaking, you still spend 90% of your time coding standard SaaS platform BS (auth, orgs, projects, etc). how is this not a solved problem by now?
English
"We classified entire branches of physics during the Cold War"
What the fuck? How is this possible?
x.com/i/status/18683…
Elon Musk@elonmusk
English
@SCPantera @eigenrobot I think they're all *compelling*, in the sense of having a reality distortion field of sorts. Attractiveness is only one way to be compelling.
English
@eigenrobot Hmmm, wonder if I can square this with universal "men are attracted to perceived femininity" theory because I agree these are unmistakably examples of femininity but some of these don't strike me as attractive
Except maybe in a fetishistic sense, which still holds
English
femininity isnt just for maidens
yes you can have a femininity performed for sex appeal but there are tons of feminine masks women can and do wear for different contexts and roles that have nothing to do with male sexual desire
eg: four feminine modes
Sampo Naatan Lähteenaro 🇫🇮@Teawisher
@eigenrobot Yea it definitely has the amending restricting part too instead of just the idealized. While performed femininity is just the fantasy and lacks for example the practical mom stuff that used to be a part of the more rounded ideal view of women.
English
@nosilverv I'm reminded of David Graeber: "Almost never do superheroes make, create, or build anything. The villains, in contrast, are endlessly creative. They are full of plans and projects and ideas." thenewinquiry.com/super-position/
English
This is nearly necessarily the case: there's a level of evilness that can't be reached unless you're very deeply and profoundly twisted and how deeply and profoundly twisted you are is bottlenecked on how much inside your mind you live: the natural world provides natural limits.
Adrian@blagojevism
One of the most disturbing revelations of adulthood is learning that the nerds (my people) are generally more evil than the jocks.
English
@alex_here_now I dunno, I always thought this one seemed pretty useful x.com/ImogenWK/statu…
Imogen West-Knights@ImogenWK
heard last night about someone who went on one of those ayahuasca retreats where they prepare you to have an incredible spiritual communion with the universe or whatever, and when he finally took the drug the vision he saw was of his own car telling him to take better care of it
English
Psychedelics don’t actually deliver any insight, they just make you feel insightful
Anyways here’s a durable insight I came to about valence while feeling insightful on psychedelics
But SERIOUSLY… I mean it!!!… psychedelics don’t actually deliver any insight
Samuel Hammond 🦉@hamandcheese
I did 4 cups of ayahuasca once and tunneled through my consciousness in a state of euphoria having one epiphany after another. At some point, the sense of epiphany continued without any meaningful referent, as if I was leaning on the epiphany key in my brain. That then triggered a meta-epiphany that psychedelics only make you feel like you're learning something new or deep, sorta like how déjà vu is probably just your brain's familiarity circuit misfiring in a context that isn't actually familiar. If psychedelics have given me any durable insight, it's that our consciousness is extremely fallible and often misleading, particularly with respect to valence.
English
@ellegist seeing someone pull that exact move was an eye-opener for me
English
@ellegist even worse, if anyone tries to hold you to higher standards then you can indignantly point out that you never consented to those standards, so how dare you etc. etc.
English
@elonmusk How much would it cost to just buy Novo Nordisk and sell Wegovy at ~cost to Americans?
English
Nothing would do more to improve the health, lifespan and quality of life for Americans than making GLP inhibitors super low cost to the public.
Nothing else is even close.
Jeremy Kauffman 🦔🌲🌕@jeremykauffman
Luigi Mangione's manifesto includes a superficially damning claim. However, it's wrong. Want to understand why the US spends so much on healthcare to get such poor results? Read this thread to see the greatest (and most unknown) analysis of US healthcare ever done.
English
@isabelunraveled @Connorpoleary I think Dawkins would happily accept that ideas and stories are real, and interact with genetics in all kinds of ways - he invented the concept of memes to say exactly that. but Peterson can't seem to distinguish between "these things are related" and "these are all one thing"
English
@Connorpoleary my impression is peterson has the humility to not speak in absolute terms; he can’t say this or that absolutely did or didn’t happen — his focus is on extracting meaning from these stories, whether they are literal or figurative, that can be acted upon to improve us as a species
English
English
@ArtirKel saints::deities
priests::gurus
rosary::mala
rose windows:mandalas
Latviešu
tpot predictions 2025
1. catholicism (it's basically vajrayana guys)
2. ayurveda
3. monogamy
Indonesia
@SailerRipley @zack_overflow Also here to say Programming Erlang. Completely changed how I thought about long-lived applications and recovery from failure modes.
English
@zack_overflow Philosophy of Software Design (Ousterhout)
Programming Erlang (Armstrong)
Thinking Forth (Brodie)
Parsing Techniques (Grune)
Advanced Programming in the Unix Environment (Stevens & Rago)
ditto SICP
English
@MatjazLeonardis Decisions reduce cognitive load, high cognitive load induces depression due to physical energy requirement?
Depression is often analogized to carrying a heavy weight; decision is to cut something away.
English
Why is indecisiveness so often depression inducing?
English