Lydia (in SF)

@dernhelm7 REAL

@LydNot Ahhhh haha yea The specific thing it was reacting to was my life goal of having a bathtub standing one line away in a lost from "meta-questions about legible impact" XD

"Very few people's abstractions can survive contact with their own groceries." generated by Fable 5 in analysis of my private corpus of written works

@dernhelm7 ahh i misread it as "it's ok that ur abstractions don't survive contact with your groceries -- don't worry -- most people's don't" lmao

@LydNot First level context: I found the sentence amusing/unexpected. It knows I don't want it to flatter me, agonizes about how not to in the exposed thinking process, and this is still flattery...but it's not the type of flattery I'd expect XD

@const_ what were the outcomes if any so far?

anyone else tried manually writing down their own CoT to finetune a model on it?

@celestepoasts they’d have to ‘get it right first time’ i guess? there’re like 50yrs / 5.5 billion people of headroom above us to test longevity treatments out on & that’s not so many getting it right first time with high confidence seems superhuman / ASI rather than AGI

I think believing we will build AGI in the next decade and also believing you won't live for 300 years if you're 20 now is pretty incoherent? Arguments against?

@PradyuPrasad may you be back

bittersweet feeling at ending my year at elicit. much much more to be done

i read ideas to try and change how i think for the better; not many ideas feel new / can make much of a dent; i’m quite acutely aware of opportunity cost of not seeing / internalizing the right ideas quickly enough LLMs crowdsource / aggregate knowledge / wisdom & make it digestible; that’s a pretty good deal if you take measures to stay ‘yourself’ / differentiated (ingesting alongside other data sources) you’d expect a perfect AI’s writing quality to be lower-bounded by the best human writing because the AI, having seen all human writing, can just reproduce extant human writing verbatim when it’s best for the task i want to be well-adapted for a world 5yrs from now; i take ai writing seriously

@LydNot interesting! concretely why do you worry more about false negatives here? or what bad outcome would it result in for you

Excited to have worked on this! AI detectors will continue to be super important in the slop-filled world we're heading towards. Overall super happy about the low FPR, and some takeaways in the thread: (1/n)

@enthusednotebk yeah who wrote it just doesn’t concern me; what it says does. surprised other ppl esp ppl ‘woker’ than me don’t think this way

i find myself not wanting to discriminate against LLMs with good ideas — ig i’ve always been more concerned about false negatives than false positives in this context — maybe bc i skim very quickly… and a lot of human writing concerns me ! seems overly preoccupied with / lingering on the ‘wrong’ themes, etc

@01shuklabhay @conduitpbc unexpected things that are conduit fashion merch spinoff arcs

@LydNot @conduitpbc my neural data --> 'fable make cool designs with this nondescript data. use images from web search.' --> 80-100 design rejection passes --> fourthwall for print for example the butterfly is an asset w/ displacement map using laplacian of my eeg

learning how to make my own clothes. less really is more!

@celestepoasts @OITIX “give me liberty or give me death!” “he who trades freedom for safety deserves neither”

@celestepoasts @OITIX

@LydNot @OITIX yeah I agree with this sadly I don't care

@celestepoasts @OITIX alternatively, “bring down the infrastructure now while we can bear it — we need to expose its shortcomings before it’s too late. we’ll build back antifragile”

@celestepoasts @OITIX i think it’s easier to steelman open-source advocacy in non-utilitarian frameworks! ones that invoke dignity and liberty autonomy sovereignty & a touch of cyberpunk

@OITIX you may presuppose utilitarianism

Can anyone steelman this opinion to me

i’ve been filtering a lot of outbound (slack messages etc) through LLMs recently. but then i thought about various characters in Public Life / the town square whom i really appreciate. almost certainly LLMs would’ve flattened, smoothed, turned back, rejected so much of their tone as they developed it 🤔 on with the schizo-posting!

@norvid_studies @croissanthology 🥅

@LydNot @croissanthology you want to know the real losers match? it's the match between you and your "recent posts" tab. @croissanthology

all might not be lost for alignment researchers: people are getting SAE work done before the argentina-england game