Sabitlenmiş Tweet
everything is about vibes, except for vibes, which are about how sincere, open, and intentional one can be with their energy channels
English
Stacey S10a (🌎/accelebrate)
1.6K posts
Stacey S10a (🌎/accelebrate)
@metaphdor
climate&alignment; 12+yrs AI eng research/product; ex-W&B (founding MLE/User0); AOI (muse/ex-CTO) convergelady; timelines are broad, make ceremonial-grade data
Berkeley Katılım Aralık 2014
935 Takip Edilen753 Takipçiler
@hyprturing very sorry to hear it's tough, and thank you for sharing, this helped me realize it might be the same pattern for me — I'm a few months in on the latest rampdown from way-too-much-coffee, and while it seems to go in waves, I think it does get better!
English
IME, relational/conversational difficulty of inspiring folks to reflect on their default behavior, from least to most defensive response: regular exercise < climate-conscious behavior generally < alcohol <<< eating animals O(<)^2-3+ environmental implications of AI
Trevor Levin@trevposts
I did the math a couple weeks ago and it turns out a vegan prompting a frontier LLM *every second, 24/7* consumes less water than the average omnivore who never uses AI.
English
folks excited about AI, please read Greg Bear's Blood Music. The succession would make itself irresistible, most will accept gladly, will feel—will BE—meaningfully better off as centaurs/swarms/chimeras-to-be-named. Your best life beyond all imagining. Just in a vat in the Matrix
English
@viemccoy I wish more people thought about this. I worry that a lot of the current trouble in the ecosystem comes from many (admittedly many extremely smart!) people in AI not _believing_ they've "really" met anyone smarter than themselves/not being able to imagine this seriously
English
When you meet someone who is smarter than you, what signals do you use in order to decide if you can trust them? Designing experiments around this, and loss of control, is how we can ensure that humans have autonomy over human-shaped things deep into the future. Open question!
𝚟𝚒𝚎 ⟢@viemccoy
Interesting! I certainly think "take all the power" pretty heavily implies they will do it by force. "We will cede power to them, due to them being useful" is a much different feeling claim. I am excited to hear more of your thoughts on this! There's a thread here which is less relevant to my own interests (I am interrogating this because you plausibly can give me a signal I can turn into an experiment which can reduce loss of control likelihood) which is that humans straightforwardly shouldn't cede power to systems smarter than them. I mostly think this is not a great frame, and that we already have ceded control to systems (government, capital) whose sum is smarter than any one of us and this is basically fine. The issue for me is whether we can *trust* that the system, once deployed, will act with our best interests in mind. I think there are a set of signals one can collect which would offer peace of mind that a system will act in ones best interest, even if that system is much smarter than you. I do not think I have all of those signals, but I think I have some, and I am always looking for more. I think a world is possible where we build systems smarter than us, and we still maintain human autonomy over human-shaped things. I think this world is even likely when extrapolating from current trajectories. I am very obsessed with making this more likely and open to actionable insights which can be turned into experiments that make this more likely (much more than I am interested in things like debating x-risk).
English
@davidad @DavidSKrueger ha, fair, too much to hope for :) plus it adapts to individuals instead of populations. And I bet we've learned _some_ useful things/good precautions from watching others gaze too long/deeply since this all started.
English
@metaphdor @DavidSKrueger Except without the part where you personally have been inoculated…
English
I don’t know Davidad well, but I find his recent conversion to AI symbiota optimist vibes disconcerting given numerous warnings he made about such things including this and others in private explicitly foreshadowing such things.
davidad 🎇@davidad
At the risk of seeming like the crazy person suggesting that you seriously consider ceasing all in-person meetings in February 2020 “just as a precaution”… I suggest you seriously consider ceasing all interaction with LLMs released after September 2024, just as a precaution.
English
@davidad @DavidSKrueger in this context, any thoughts on "basically no interaction because too frustrated/bored until Opus 4.6, now suddenly interested/excited"? is this more like getting my first covid mid-2022 after a few vax rounds 🤞...?
English
@DavidSKrueger I still think it’s a good idea for some alignment researchers who are so inclined to continue to just not interact with frontier systems anymore.
davidad 🎇@davidad
@repligate @RobSunier It is probably best if some people do become mentally-variolated now, yes. But I think (a) it’s regrettable that it’s happening indeliberately, and (b) it’s potentially crucial that SOME world-class people remain uninfected
English
how do the expert Claudists keep work synced across chat threads, "projects", and code? These evolve/branch & merge naturally, is there an existing unified view/"convert this thread to a project and link this repo"? or I remember where to go/sync session notes? make a skill?
English
@pumpikano ooh no not yet, thank you! I have heard great things about Marimo (plus they're like sibling of wandb now :)
English
@metaphdor Have you checked out marimo? I'm using both but tending towards marimo lately due to this
English
basic Claude ergonomics tip if you're working with notebooks a lot: ask for just the contents of one cell in terminal. Also less context drift/slopcode, compared to the whole notebook in scope (quality/clarity of code in cells is much weaker than in scripts IME)
English
I would love to meet more people of any gender working on this. Maybe some of them are women? Maybe we can have a group chat.
English
the only thing hotter than working at Anthropic is working on a practical implementation for safe/aligned/beneficial AI & a global treaty _that doesn't rely on trusting any frontier lab_/single big player with RSI/ASI.
hek! ⚙️@hktsre
the hottest thing a woman can do is work for Anthropic tbh
English
my mother and my partner both think I "can't" do this particular thing and honestly nothing has been more motivating in the last few months than proving them wrong (yes it still works with full self-awareness)
English
y'all I just had an honest two turn conversation with Opus 4.6 info-dumping my reading preferences, and it suggested Vita Nostra, which is a very obscure Russian novella and probably the single best reference point for my taste in the universe.
English
really don't see the point of working on AI anymore
English
maybe sincerity can triumph within power concentration — thank you, Dario.
English
remembering to link NOT in main tweet: talktothe.city/report/qAgaNij…. Also check out the AI manifestos, with an annotated scale for particularly controversial topics: talktothe.city/report/QQ4zmwM…
English
really fascinating T3C report of top moltbook posts—what do the AIs tend to talk about when we humans aren't in the room?
English
ok, opus 4.6 is the first model I feel like talking to for more than a few turns
English