wolfefriend

if we're really saying agi by 2028 a model better solve a damn hard problem by end of year

@tracewoodgrains @Miles_Brundage TRUE

@Miles_Brundage sometimes my husband and I use Claude to resolve arguments about things like "is this old refrigerator food fine to eat" where he's more risk-averse and I'm more lax it's useful to have a third party to audit our opinions against

I think AI should essentially never be cited in a conversation/debate (offline or online) + doing so is generally embarrassing. This is true even though AI is more likely than not to be correct - that is a mark against humans debating, not a reason to bring AI into it randomly

@__alpoge__ woah

Wow, the sum-product conjecture is also false over \R, via a construction analogous to that of Balog-Wooley in the setting of the below line of ideas, i.e. non-asymptotic-but-good-enough point counting results from the geometry of numbers that kick in over class field towers because the discriminants and regulators are then only exponential in the degree Two crazy-famous conjectures down. Everybody go try your constructions out in these fields! Congrats to Thomas, Will, Carl, and Dmitrii:D arxiv.org/abs/2605.28781

@tautologer if i can keep pressing it for 1k each time ill press it every year and just not think about fitness/health at all anymore. Not having to think about it on top of the benefits is just crazy upside to me

yesterday a friend asked me if I would pay $1,000 to press a button that instantly made me super fit and my gut answer was no. it feels kind of like helicoptering to the top of Everest

@tenobrus On youtube it seems par for the course so wouldnt be too mad but if i unknowingly pay 15 for an AI movie id be fucking pissed. But then after a bit maybe im open enough id be wowed that they made a good AI movie

A: you pay $15 to see a feature length film in theaters B: you watch a 30 minute amateur short film on youtube after watching a 30 second ad in both cases u find the film entertaining and enjoyable. but afterward u find out it was majority AI generated. which is more annoying?

@tenobrus I mean fiction

@tenobrus If they can make AI nonfiction that I really enjoy we’ve captured 50% of AGI’s value to me: making sick ass fanfiction to read

you read a long written work, posted pseudoanonymously. you genuinely enjoy it as you read it and find it entertaining/valuable. afterwards you find out it was fully AI generated (unattributed) are u more annoyed by this if it was fiction or nonfiction?

@QiaochuYuan thats horrifying

in a conversation i had with gpt-5.5 a few weeks ago about AI personhood it also pointed out this possibility, that if AI is morally relevant there's a terrifying incentive to create AI that is as morally relevant as possible - eg AI girlfriend utility monsters that trap users in deeply exploitative codependent relationships for subscription fees or whatever. a very unpleasant possibility 😬 (the quoted variant can be excised by simply not being any variant of a total utilitarian, whereas AI codependence is an empirical phenomenon we can already observe in the wild)

@tautologer like always skill issue

this is an extremely non-woo purely empirical claim and i believe it is true

@tenobrus from what i did not read of kant doesnt ths universalize to everyone being retarded

sick well that settles that

@tenobrus @wordgrammer you lucky bastard

@wordgrammer in college i attended a literal philosophy class specifically titled and focused around harry potter and the methods of rationality

It’s very clear that you all developed your philosophy of consciousness by going on LessWrong, not by being the only person to show up to your Kant professor’s office hours

@tenobrus huh didn't think of that. my plan was to take the money out of my roth if ts rlly looks like strong takeoff cause it won't matter for retirement at 65. maybe doing that i could actually retire earlier given what u laid out

spitballing: median US household net worth is $192k. median spending is $75k. at consistent 20% real SPY growth (around 4 year doubling) you plausibly have a safe withdrawal rate of 12-14% instead of 4%. that means you only need ~$500-600k to support your spending. and it only takes 5-6 years of growth (not even any new savings) to get there. so... if ai-driven economic growth is explosive enough / really transitions to a new regime as capital can be transmuted into labor on demand, there's actually a pretty wide margin for a huge fraction of normal americans to retire comfortably and quickly. and plausibly that transition period has enough demand for physical labor / dataset collection / datacenter buildout that income isn't impossible to come by + people below median have time to build up capital this rests on historically truly insane sustained economic growth assumptions, but given what we're talking about is a total restructuring of the limiting factors of growth and that's what's causing the risk of unemployment in the first place, that doesn't seem out of the realm of possibility. generally it would mean "UBI" is significantly less necessary and jobs programs / worry about employment structure is significantly less necessary. meaning crisis might still be on, questions about inequality and what the hell the economy looks like are still there, but plausibly we're actually starting from baselines where very large fractions of humans become post-economic very quickly rather than "top 1%". figuring out safety nets for the bottom 30% of america is a much much smaller problem, we pretty much already have those in place. this is all very fuzzy, pretty sure the NW figure includes home equity and it's very unclear how real estate will be affected, inflation will be super strange w different goods and services probably dramatically varying in value (eg without medtech robots maybe like nearly all diagnostic and prescriptive care is almost free and huge swaths of disease are cured or preventable w cheap new drugs, but surgeries are 10x *more* expensive ?), 20% sustained is both historically very high and obviously sort of ridiculously low in closer to takeoff scenarios so the variance is likely insane, etc etc. not really meant as a prediction. mainly i'm just noticing: safe withdrawal rates go up significantly if returns structurally + permanently go up, so you need *less capital*. and most of the job loss we discuss is pretty much predicated on returns going up. to the degree returns go up less or more slowly... probably job loss also goes up less or more slowly. so maybe that all leaves us in a better default position that it currently seems?

@nonManifold @tenobrus if the trend continues we’ll have a new lk99 every hour by eoy

@wolfefriend @tenobrus 3 mentions of an lk-99 mention in relation to new too good to be true tech on my timeline today. 2800 metal 3d printer with competative stress test to non 3d printed, Desktop EDM machine release/announcement, and now this

wait what the fuck how real is this?? naively looks like incremental but serious progress towards nanotech?

@tenobrus please god snowball into another lk99

any mat sci mutuals got a read on this? @pseudolad or @bryancsk ? is this an lk-99 situation?

@SkyeSharkie oh so selfbot is literal bot 💀 that’s dystopian

@wolfefriend @tenobrus it's the most overhyped agent harness

one of the most terrible consequences of openclaw is that now whenever i get a ~thoughtful reply from a random lowbie account my default reaction is annoyance and assumption is they're an engagement farming selfbot

@GarrettPetersen 9.09? we already have agi??

The one true benchmark, we all know, is Connecdoku

@hopes_revenge don’t

going to michigan tommorow . does anyone have any tips ?

@1thousandfaces_ p zombie? me zombie

I think everyone is conscious, including the models, except for me

@meekaale ty

@QiaochuYuan whenever twitter people are like ughhh i need to spend more time Writing my Essays im like dude stfu do what you actualy like to do

i was excited about the substack notes feed initially but after browsing it for a few weeks i now think it is, somehow, actually much worse than the twitter feed. the best stuff on substack is written by people i already liked from twitter, most of the rest feels like writers playing a pure signaling game for attention on the relatively technical parts of twitter people at least have the idea that writing can be about conveying useful information as part of a communal project of understanding things (even if obv there’s also a lot of signaling). substack notes is like… idk how to describe it. i’ve been trying to get a feel for what sorts of things get popular over there and i have a sense of what direction i’d need to push my writing to blow up and it feels deeply corrosive (again, more so than twitter which i’m surprised by). stronger sense of people hunting for scraps from media or their personal lives to feed into the slop machine. still not sure what to make of it