Aaron Bergman's tweets, September 2024
Every public tweet, reply, quote tweet and retweet that Aaron Bergman (@AaronBergman18 on X, formerly Twitter) posted in September 2024 and has not deleted: 303 posts, oldest first, times in UTC. Each heading links to the original post on X. When Aaron replies to or quotes someone, their tweet appears as a blockquote with their name; those words are theirs, and everything outside the blockquotes is Aaron's.
This page is one month of the complete archive of Aaron Bergman's tweets, which updates automatically every few hours. Previous month: August 2024. Next month: October 2024.
Sunday, September 1, 2024
Reply to @slimepriestess, 03:37 UTC · 1 like
Ra (@slimepriestess): okay apologies for my bullshit, im gonna try and be less vulnerable and mentally ill on the TL. maybe i need a priv actually.
That’s what twitter is for
Tweet, 07:58 UTC · 12 likes
I can’t believe that computers work at all
Thread, continued, 08:02 UTC · 5 likes
Honestly the incandescent lightbulb is at the edge of my comprehension
Anything invented in the 20th century or later is magic
Reply to @RafaRuizdeLira, 08:49 UTC · 4 likes
Rafael Ruiz ⏸️🔸 (@RafaRuizdeLira): @AaronBergman18 Took two million ape years and a hundred billion naked apes to come up with this shit
I wonder what the apes are up to now
Reply to @liz_love_lace, 10:00 UTC · 5 likes
Liz Lovelace (@liz_love_lace): update: they went "actually, wtf have you been doing with our servers? this does not look like normal usage, if you continue to fuck around we might ban you"
(quoting @liz_love_lace: opening tickets with my server provider makes me feel so grown up and serious. "Yes I'm an important business client and I need you to raise my maximum server instance cap, we require it for our operations")
For the curious
Reply to @lastpositivist, 10:10 UTC · 5 likes
Liam Bright (@lastpositivist): @NathanpmYoung That thing scares me
Dw, pleased to announce I figured it out https://www.aaronbergman.net/p/the-answer-to-the-sleeping-beauty?utm_campaign=post&utm_medium=web
Quote tweet, 10:24 UTC · 3 likes
The correct wise centrist position is that straightforward image (or whatever) generation models are meaningfully different from human art for the reasons Chaing describes (idc if you wanna use the word “art”) but this doesn’t imply that AI can’t make art
Quoting The New Yorker (@NewYorker): In a new Weekend Essay, Ted Chiang argues that artificial intelligence can’t make real art. https://newyorkermag.visitlink.me/-UQFZr
Tweet, 10:29 UTC · 360 likes
I took one good picture with a good haircut and now I have probably 3x the number of hinge matches
What the fuck
Thread, continued, 10:29 UTC · 33 likes
If you found this from Hinge somehow before I delete, hello and I hope you’re having a nice day
Reply to @MikeFHay, 10:39 UTC · 45 likes
Michael Hay 🇬🇧 ⏸️ (@MikeFHay): Show pic
Thread, continued, 10:39 UTC · 15 likes
Just kidding lol. Technically 2 pics
Reply, 17:08 UTC · 20 likes
Reply to @h3xenbrenner2, 22:33 UTC · 1 like
hb (@h3xenbrenner2): there should be something called "utility sauce" which goes with every single food. Chicken, pizza, ice cream, curry, yoghurt, ribs, cous cous, stew, whatever
MSG
Tweet, 23:27 UTC · 7 likes
Implicit in LLM models is the existence of a most shocking set of n tokens: the least likely token given no prior context, followed by the least likely token conditional on that and so on
What do we think it is? Would it be qualitatively interesting?
Monday, September 2, 2024
Reply to @NoahTopper, 11:26 UTC · 1 like
Quinoah 🔍⏸️ (@NoahTopper): Have you considered looking happier in your photos
Lol the purpose of these was to make a direct comparison before vs after but yes perhaps a smile or perhaps a smirk would be wise
Reply, 11:36 UTC · 2 likes
Sweet, I will continue going to the barber every other month and occasionally wearing this shirt
Reply to @SpacedOutMatt, 11:41 UTC · 1 like
Matt 🔸 (@SpacedOutMatt): @AaronBergman18 @VickiRose95 U should go to Hamid monthly
I could become a barbershop guy
Tweet, 18:43 UTC · 6 likes
Cmon lol
Tuesday, September 3, 2024
Tweet, 01:42 UTC · 9 likes
I selfishly wish real money prediction markets were easier to use for small scale stuff like Manifold bc I’d love to hedge against things that would personally suck (i.e., if they do, I get some money as consolation)
Thread, continued, 01:42 UTC · 2 likes
This would be free money for everyone else (in expectation), just like with normal stuff insurance (car, home, life)
Reply to his own earlier tweet, 03:16 UTC · 21 likes
Aaron Bergman, September 1, 2024: Implicit in LLM models is the existence of a most shocking set of n tokens: the least likely token given no prior context, followed by the least likely token…
Thanks to @livgorton, we now have a result for Llama3-8b
It is not especially interesting
Reply to @bashu_thanks, 03:18 UTC · 2 likes
bashu, thanks (@bashu_thanks): @AaronBergman18 i don't understand this one. can you explain it
1-p-lsd is a grey market molecule that metabolizes into lsd and is maybe not technically illegal
There are no other layers here, it’s not especially clever lol
Reply to @thricewrithed, 04:30 UTC · 1 like
niko (@thricewrithed): telling my therapist i live in constant mental anguish and he says he doesnt see what the problem is because my grades are good. “it’s not impacting your life” DIE DIE DIE DIE
U should find a new guy
Wednesday, September 4, 2024
Reply to @livgorton, 03:36 UTC · 2 likes
Liv (@livgorton): wild how a little gremlin who cries all the time, deprives me of sleep, steals my socks, and won’t let me have time to myself for more than a few hours has been the best thing i’ve done for my happiness.
Dog 🍀
Thread, continued, 03:46 UTC · 2 likes
Also photogenic dog
Quote tweet, 04:36 UTC · 4 likes
Build the cl-1201
Quoting Luke Leisher (@Lukeleisher): One day I hope I have enough followers so that I can type “we should have built the cl-1201” and 100 people will reply “real”
Reply to @albrgr, 04:49 UTC · 3 likes
(The tweet being replied to by @albrgr is no longer available.)
Reply to @horrible_bug, 05:24 UTC · 2 likes
(The tweet being replied to by @horrible_bug is no longer available.)
Absolute unit of a bird
Reply to @KatSpartz, 20:17 UTC · 4 likes
Kat Spartz ⏸️ 🔶 (@KatSpartz): Try asking ChatGPT who it thinks you are.
Share the results in the comments.
For me, ChatGPT knew me crazy well.
Kinda made me tear up a little. 🥹 I felt really seen.
Similar:
Reply to @AlecStapp, 21:03 UTC · 2 likes
Alec Stapp (@AlecStapp): Weird situation where firefighters (mostly) don’t fight fires anymore
Yeah in hindsight when this happened it’s a little weird that like 4 fire trucks showed up
Quote tweet, 22:02 UTC · 10 likes
This is certainly a tweet
Quoting New York Magazine (@NYMag): Mark Robinson, the Republican candidate for governor in North Carolina, reportedly watched porn in private booths up to five days a week in the ’90s. He strenuously denies that as well as the Holocaust. https://trib.al/Lt0PJd8
Tweet, 22:55 UTC · 17 likes
Edge of Sentience is such a banger. If you’re at all interested in altruism or ethics (or at least enough to read other books from this motivation) you should probably read https://www.edgeofsentience.com
Thursday, September 5, 2024
Reply to @TylerAlterman, 00:11 UTC · 10 likes
Tyler is finishing a book, slow to reply (@TylerAlterman): Think about the most positively life-changing 30-minute-or-less event that has ever happened to you. What was it?
(Other than birth events or meeting the love of your life)
Successful surgical removal of a chopstick from inside my neck
Reply, 00:19 UTC · 1 like
Not that great of a story lol
Tweet, 02:09 UTC · 5 likes
Decontextualized, IRBs actually sound like a good idea for avoiding unnecessary suffering (“is there a way to do this research that doesn’t involve torturing 100 mice to death”?)
Too bad afaik they don’t give a fuck about this
Reply to @arithmoquine, 02:38 UTC · 1 like
henry (@arithmoquine): @AaronBergman18 idk i disagree based on nothing more concrete than intuition about myself
Solid, you know urself much better than I do lol
Tweet, 04:30 UTC · 7 likes
HOLY SHIT I FINALLY MADE SOMETHING WITH A BACKEND (ie the votes are cumulative among all y’all) https://hidden-sierra-87210-06f8117ea5fb.herokuapp.com/
Reply to @eshear, 05:10 UTC · 2 likes
Emmett Shear (@eshear): I think SB 1047 has gotten much better from where it started. It no longer appears actively bad.
But can someone who is pro-SB 1047 explain the specific chain of causal events where they think this bill becoming law results in an actual safer world? What’s the theory?
This is specifically not what you asked but there’s also an evidential decision theory case for it
The worlds where SB 1047 passes look better from a non-causal standpoint (evidence about willingness to pass AI safety legislation at least)
Reply to @BearPosting, 12:27 UTC · 1 like
🐻 Bear 🎂🍂 (@BearPosting):
@RafaRuizdeLira 🐻
Reply to @SpencrGreenberg, 17:02 UTC · 25 likes
Spencer Greenberg 🔍 (@SpencrGreenberg): A question for you: what's the worst thing a person can do that isn't illegal?
Question by Seemay Chou (@seemaychou)
I have ideas that are way more ambitious than the ones given here, but they’re not worth a .01% risk of putting in someone’s head
Tweet, 18:39 UTC · 8 likes
Wanted: @birchlse review of @RethinkPriors moral weights/welfare ranges project https://rethinkpriorities.org/publications/welfare-range-estimates
Reply to @pearl_please, 19:50 UTC · 4 likes
(The tweet being replied to by @pearl_please is no longer available.)
Take: people who have strong opinions on value drift vis a vis EA are the people who’s values are least likely to drift
Friday, September 6, 2024
Reply to @PeterJAngel, 01:14 UTC · 8 likes
Peter🧅📈 (@PeterJAngel): Is there anything I can do in the next 2 hours that will promote the good, and human flourishing?
Apply for 80k coaching if you haven’t
Admittedly they’re running a referral link thing but you should do it regardless
https://80000hours.org/speak-with-us/apply/?referred_by=Aaron Bergman&referrer_Id=1262602471
Quote tweet, 01:18 UTC · 27 likes
If we’re mutual and you haven’t applied for @80000Hours coaching, you probably should
I have a referral link (below) that could result in me being given money that you should use (which is the proximate reason for this tweet) but I really do think it’s (probably) wise to apply
Quoting Aaron Bergman 🔍 (@AaronBergman18): @PeterJAngel Apply for 80k coaching if you haven’t
Admittedly they’re running a referral link thing but you should do it regardless
https://80000hours.org/speak-with-us/apply/?referred_by=Aaron Bergman&referrer_Id=1262602471
Thread, continued, 01:18 UTC · 2 likes
https://80000hours.org/speak-with-us/apply/?referred_by=Aaron Bergman&referrer_Id=1262602471
Reply to @RichardHanania, 04:19 UTC · 2 likes
Richard Hanania (@RichardHanania): Testosterone prescriptions going through the roof. Should most men be on it?
Idk about most, though definitely more. I was on TRT for a few years - exogenous hormones are finicky to get right, and more is definitely not better (even within the standard reference ranges, say <1200 ng/dl)
Probably not worthwhile for almost anyone at >500 natural
Reply to @NoahTopper, 05:59 UTC · 14 likes
Quinoah 🔍⏸️ (@NoahTopper): Neel Nanda is in his 20s, it’s so over for me
(quoting @NeelNanda5: @notanastronomer I'm in my 20s and do this, just seems rude not to!)
Infohazard
Tweet, 09:50 UTC · 2 likes
EtG startup incubator - looks incredible my based. Apply by the 15th
https://www.aimfoundingtogive.com
Reply to @NeelNanda5, 11:36 UTC · 22 likes
Neel Nanda (@NeelNanda5): I'm 25 lol
Wtf
Reply to @elvisnavah, 12:51 UTC · 1 like
Elvis Nava (@elvisnavah): I thought you were like 35
Same basically. Maybe like 32
Reply to @albrgr, 14:50 UTC · 9 likes
(The tweet being replied to by @albrgr is no longer available.)
Based
Reply to @absurdlymax, 14:50 UTC · 2 likes
Max Alexander (@absurdlymax): @JesseTayRiver That dog has a knife
Real
Saturday, September 7, 2024
Quote tweet, 09:38 UTC · 8 likes
This is good and does a great job explaining terms that smart people throw around (“sparse autoencoder” is my favorite, smth about the syllables)
Quoting sarah (@littIeramblings): I recently wrote an introduction to mechanistic interpretability for @BlueDotImpact’s AI Safety Fundamentals curriculum! If you’re looking for a high-level explainer of what mech interp is & what progress has been made so far, you may find it useful (link below)
Quote tweet, 11:36 UTC · 3 likes
Sarah kept up (as you will hear)
Quoting sarah (@littIeramblings): teaser for my next podcast episode, once again featuring @absurdlymax and @AaronBergman18 arguing about philosophy while I just barely keep up
dropping tomorrow (probably)
Tweet, 11:38 UTC · 23 likes
I did a programming thing Claude sonnet couldn’t do with multiple tries (maybe first time?)
Plz clap
Thread, continued, 11:39 UTC · 4 likes
It was in R, I’m guessing it would have gotten it right in python but I will choose to live in ignorance
Reply to @WillowChem, 11:58 UTC · 16 likes
MagpAI (@WillowChem): What's the most cruelty free way to easily consume 20g of protein in the morning, with minimal carbs, that isn't tofu?
I mean if this is ur only constraint there are plenty of vegan protein powders
But fwiw I’d say drink whey and donate some small amount to effective animal charities and you’re in the green
Reply to @datepsych, 12:23 UTC · 6 likes
(The tweet being replied to by @datepsych is no longer available.)
It’s funny that “social justice” has become a pejorative (for reasons I understand tbc) bc at decontextualized face value the phrase seems to be pointing to a good thing
Reply to @kartographien, 12:25 UTC
Kart ographien (@kartographien): if i was asked to invent an RL algorithm in 1950, I would've one-shotted PPO
Tweet, 12:31 UTC · 4 likes
I need to figure out how to make the hinge push notification sound distinct from any other push notifications bc I’ve been reduced to a dopaminergic Pavlovian slot machine addict every time I get a calendar reminder or smth
Reply to @kartographien, 12:31 UTC
Kart ographien (@kartographien): tbf gradient descent was invented by dozens people, and it's reinvented pretty much every time someone faces a problem best solved with gradient descent but they haven't heard of it
😢
Tweet, 12:53 UTC · 3 likes
I will never ever pay rent before the last day I can contractually get away with it bc goddamnit I am going to be collecting my 76.4 cents of interest (yes that’s the real amount), not some landlord
Tweet, 13:46 UTC · 2 likes
Claude says that, after redundancy and error correction, the rate of errors in ML training is about 10^-21
So, during all this (which probably took something like 10^25 FLOP), only ~10k actual computations ended up wrong
Tweet, 13:47 UTC · 1 like
^
10^-21 errors per floating poiny operation, or one error every 10^21 operations
Thread, continued, 13:49 UTC · 2 likes
A thousand times more accurate
Reply to @absurdlymax, 14:34 UTC · 2 likes
Max Alexander (@absurdlymax): I need to go to rehab for my 'or something' addiction
(quoting @littIeramblings: teaser for my next podcast episode, once again featuring @absurdlymax and @AaronBergman18 arguing about philosophy while I just barely keep up dropping tomorrow (probably))
No babe ur perfect
Reply to @NoahTopper, 15:41 UTC · 4 likes
Quinoah 🔍⏸️ (@NoahTopper): @littIeramblings @absurdlymax @AaronBergman18 aaron clearly won this exchange
I happen to agree
Reply to @Aizazadi_, 15:43 UTC · 3 likes
Aizazadi (@Aizazadi_): @littIeramblings @absurdlymax @AaronBergman18 (what's the name of ur podcast again)
Consistently Candid (Sarah’s) https://podcasts.apple.com/us/podcast/consistently-candid/id1732326178
Reply to @NoahTopper, 15:54 UTC · 1 like
Quinoah 🔍⏸️ (@NoahTopper):
I don’t actually understand backprop so if that’s not the most natural way to on-ramp from multiple regression I don’t think I’d hit on it…
Tweet, 16:12 UTC · 6 likes
Spotted in Arlington, VA
Turns out “save our tree canopy” was a tagline from a 2022 campaign to block *checks notes* allowing townhouses in formerly single family house-only zoning
Thread, continued, 16:12 UTC · 2 likes
The NIMBYs actually lost in 2023 (yay). A modest win but still https://www.washingtonpost.com/dc-md-va/2023/03/22/arlington-missing-middle-vote-zoning/
Reply to @NoahTopper, 17:04 UTC · 5 likes
Quinoah 🔍⏸️ (@NoahTopper): @lillybilly299 Pure @AaronBergman18
It means I care about her!
Tweet, 19:04 UTC · 2 likes
Plz consider emailing Gavin Newsom about SB1047, especially if you live in CA
The bottom section here provides an email address and more info! https://safesecureai.org/get-involved
Sunday, September 8, 2024
Reply to @MariGO2thepolls, 03:38 UTC · 2 likes
🏳️🌈Mar G-O 🌐🇺🇦 (@MariGO2thepolls): An explanation of why you feel a certain way is not an argument smh
I kinda disagree, this is what argumentation looks like if you’re interested in getting the right answer
Explain why you believe what you believe
Reply to @GjMcGowan, 09:40 UTC · 2 likes
George McGowan (@GjMcGowan): Latest Andrex puppy looks rubbish, 5/10
Don’t be mean 😡
Reply to @mealreplacer, 10:08 UTC · 2 likes
(The tweet being replied to by @mealreplacer is no longer available.)
@arithmoquine do u have any advice for a young up-and-comer?
Reply to @noetic_emetic, 19:08 UTC · 2 likes
matty matt (@noetic_emetic): Genuine question..
Did Immanuel kant effect any of you guys to become a better person?
Yes by being very clear about why one specific ethical system, Kantianism, is wrong
Reply to @thecaptain_nemo, 19:28 UTC · 4 likes
Nemo (@thecaptain_nemo): ozempic infused zynn pouches
It’s injectable for a reason, probably too low bioavailability
Reply to @boreal_country, 19:34 UTC · 2 likes
Beau (@boreal_country): @AaronBergman18 @thecaptain_nemo Zyn with microneedles
Now we’re talking
Reply to @LandscapeL26502, 19:39 UTC · 3 likes
Loss Landscape (@LandscapeL26502): lol oral bioavailability @ 0.8%
compared to sub-q @ 89%
Just gotta 100x the dose
Quote tweet, 19:41 UTC · 2 likes
I’ve thought about making this. Based
Quoting ❤️🔥 xiq (@exgenesis): announcing the community archive!
a quest to back up the tpot canon
links in replies
Reply, 19:46 UTC · 2 likes
Patience
Monday, September 9, 2024
Quote tweet, 02:12 UTC · 4 likes
Me and who
Quoting Buitengebieden (@buitengebieden): The Kagu, an almost extinct bird, meets its own kind.. 😊 https://x.com/buitengebieden/status/1832798880561340713/video/1
Reply to @daniellellecco, 02:50 UTC · 2 likes
(The tweet being replied to by @daniellellecco is no longer available.)
@AaronBergman18
Reply to @livgorton, 03:20 UTC · 2 likes
Liv (@livgorton): sometimes my code runs so quickly that i’m just convinced it has to be a mistake (me rn)
Too good for this world
Reply to @timfduffy, 03:21 UTC · 2 likes
Tim Duffy (@timfduffy): @littIeramblings @AaronBergman18 @absurdlymax I fully agree with @absurdlymax. I think the most devastating argument is the one @AaronBergman18 himself acknowledges is the biggest weakness, the threshold where you go to negative infinity. @littIeramblings your comparison of 50 people suffering x amount vs 1 person ...
😢
Tweet, 03:47 UTC · 14 likes
I am once again asking everybody to just say what you think is right without gaming out 9d-chess involving your probably-mistaken model of public opinion
This isn’t subtweeting anything salient, I’ve just been reminded by a few small things
Tweet, 20:34 UTC · 17 likes
Whoever invented coffee flavored ice cream deserves a prize
Reply, 20:38 UTC · 2 likes
Wait I assumed u mean like woke socialist
If u mean left-handed this is even funnier
Reply to @inesferhumi, 20:41 UTC · 4 likes
Inés (@inesferhumi): @AaronBergman18 @pearl_please She 100% meant left-handed
I can do this with 89% accuracy by guessing “no” for everyone
Tweet, 23:00 UTC · 8 likes
Tf is a data lake
Reply to @adam__isom, 23:03 UTC · 2 likes
(The tweet being replied to by @adam__isom is no longer available.)
There’s also a much stronger case to be made that p(heads | 9 heads) is ~1
At 100 consecutive heads, you’re not gonna convince me that p(marginal heads) is 1/2
Reply to @tescrealaccist, 23:04 UTC · 2 likes
TESCREAL/acc💙 (@tescrealaccist): @AaronBergman18 Its where data sits pre-processing but post-creation
How do I get access to a data lake
Any data lake idc what it’s about, I just wanna go swimming
Tuesday, September 10, 2024
Tweet, 01:17 UTC · 8 likes
AI wrappers are underrated, not overrated
Actually wrappers for software in general are underrated
Wrap the python/rust thing in a UI and make it usable for people
Quote tweet, 01:43 UTC · 1 like
Recent discourse has me wondering this again
Quoting Aaron Bergman 🔍 (@AaronBergman18): Does anyone have a thought experiment that demonstrates why a coherent ethical theory can’t have the Pareto principle, transitivity, and impartiality?
Bc if not and it just “falls out of the math,” I’m skeptical that the math is modeling the right thing
Reply to @focusfronting, 15:05 UTC · 33 likes
(The tweet being replied to by @focusfronting is no longer available.)
I basically agree but don’t know what to
make of it
My dubious synthesis based on something I read a while ago is basically: 70 IQ measured in people exposed to childhood education is fundamentally different than 70 IQ measured in people who received no or minimal education
Thread, continued, 15:07 UTC · 15 likes
In eg the US, virtually everyone receives primary education and so 70 IQ reflects a serious generalized cognitive deficit
Among the global south, the deficit is basically confined to abstract reasoning and people can do normal concrete cognitive tasks just fine
Reply to @GjMcGowan, 20:06 UTC · 2 likes
George McGowan (@GjMcGowan): Might get into jars
Fun
Tweet, 20:24 UTC · 5 likes
Is Google doing illegal predatory pricing with Gemini?
I have no idea what I’m talking about, but pricing Flash at ~1/4 the cost of Claude Haiku seems like a smart way to deprive a smaller competitor of capital and a very dumb way to make money directly
Wednesday, September 11, 2024
Reply to @mattyglesias, 03:32 UTC · 4 likes
(The tweet being replied to by @mattyglesias is no longer available.)
☹️
Reply to @amplifiedamp, 05:42 UTC · 10 likes
&. in sf 6th-8th (@amplifiedamp): The madman fucking did it wow
Credit where credits due
Reply to @NathanpmYoung, 12:50 UTC · 2 likes
Nathan 🔎 (@NathanpmYoung): .@pmarca funding an entry he liked in the EA community choice fundraiser is positive sum behaviour. Props to him.
(quoting @amplifiedamp: @deepfates @pmarca The madman fucking did it wow)
Reply to @seconds_0, 13:22 UTC · 83 likes
(The tweet being replied to by @seconds_0 is no longer available.)
Did u find any
Tweet, 13:43 UTC · 7 likes
Looking forward to this
Reply to @manic_pixie_agi, 13:55 UTC · 2 likes
Corinne (🏠 in NYC) (@manic_pixie_agi): https://x.com/amplifiedamp/status/1833816401137574171?t=h_EZg51p1-W_agPMR38LlA&s=19
Based
Reply to @AndyMasley, 14:57 UTC · 4 likes
Andy Masley (@AndyMasley): They are eating the birds
They're eating the pigs
They're eating the fish
😢
Reply to @DTWillingham, 15:38 UTC · 2 likes
Daniel Willingham (@DTWillingham): The brain did not evolve to perform many of the tasks we value, e.g., reading & math. We must learn these tasks by piecing together cognitive modules that evolved for other purposes. That raises the question "how many of these modules are there, what are they,
1/2
Relevant caveat is that the brain co-evolved with language, even if not reading
Tweet, 19:52 UTC · 7 likes
On the relative merits of philanthropy/donation vs veganism/ethical consumerism for animals
Reply to @MikeFHay, 21:40 UTC · 1 like
Michael Hay 🇬🇧 ⏸️ (@MikeFHay): @AaronBergman18 Surely the existing demand for alternative meat products is also a large driver of investment in clean meat and other vegetarian products? Seems inconsistent to dismiss the impact of paying customers while holding up the impact of investors.
Yeah this is a fair point, although it depends on how much profit companies are making from sales (I have no idea but my guess is not much for most things)
Also I’m thinking about nonprofits (GFI mostly) more than investors, although foregone returns via impact investment
Reply to @mealreplacer, 22:56 UTC · 2 likes
(The tweet being replied to by @mealreplacer is no longer available.)
😢
Thursday, September 12, 2024
Reply to @PrinceVogel, 00:15 UTC · 9 likes
Prince Vogelfrei 🐦⬛ (@PrinceVogel): does anyone have any compelling reasons I shouldn't try the horse electrolytes
Unironically I wouldn’t do this bc electrolytes are rocks from the frickin ground lol. I would not eat arbitrary sodium chloride (or whatever) of unknown purity not intended for human consumption
Thread, continued, 00:15 UTC · 2 likes
As opposed to say medications which are probably from a pretty well-controlled supply chain
Thread, continued, 00:17 UTC · 5 likes
But by all means make your own electrolyte mix
Iodized table salt
Potassium chloride
Magnesium whatever (not oxide)
Zinc whatever (not oxide)
Ask Claude for the ratios probably
Reply to @livgorton, 00:38 UTC · 2 likes
Liv (@livgorton): we are grouping we are theorising
image attached
(quoting @livgorton: does anyone have group theory textbook recommendations? had some related research ideas but realised i don’t really know all that much about group theory (slight barrier))
Reply to @Macbaconai, 12:41 UTC · 3 likes
Mac Baconai (@Macbaconai): please, more space under the grounds..
🤝
Quote tweet, 13:48 UTC · 6 likes
It would be cruel and unusual punishment to deprive data nerds of this data
For your convenience, here's everything available as of yesterday (takes a bit of setup and a little while to download directly) https://drive.google.com/drive/folders/1e0NMIniZ3t6MxR33ee3aKdPXj73oAuLL?usp=sharing
Quoting ❤️🔥 xiq (@exgenesis): reminder to request your twitter archive :)
Tweet, 14:31 UTC · 12 likes
Tfw you have to actually register to vote, not just identify as the kind of person who is registered to vote
Reply to @bschne, 16:08 UTC · 2 likes
Benjamin (@bschne): Her: "What's your favorite movie?"
Me: "I'm sorry, that's kind of a broad question, what do you mean exactly?"
Reminds me of when someone said they didn’t love a book I recommended and I had to clarify that I never recommended it or said it was good, just that I liked reading it
Reply to @GjMcGowan, 16:25 UTC · 3 likes
George McGowan (@GjMcGowan): Annoying how companies will email you and then be like “DON’T EMAIL BACK, THIS MAILBOX IS NOT MONITORED”
Why not? I want to email you back, this is a two way relationship you know.
Sometimes the inbox is monitored
Quote tweet, 18:54 UTC · 2 likes
Rip
Quoting Shakeel (@ShakeelHashim): I took a stab at summarising the OpenAI o1 system card. A few bits in particular jumped out at me:
1: @apolloaisafety finding the model "instrumentally faked alignment during testing", and deeming the model capable of "simple in-context scheming".
Friday, September 13, 2024
Tweet, 18:09 UTC · 9 likes
Why am I browsing ultralight aircraft on fb marketplace again 😩
Thread, continued, 18:11 UTC · 3 likes
I yearn to fly
Reply to @animalologist, 18:24 UTC · 10 likes
taco belle (@animalologist): One time a man told me I was even more cute and smart and interesting than he thought I’d be from my dating profile and when I said thank you he said “wow not even gonna be humble about it?”
(quoting @guapunzel: a lot of men like this are secretly trying to humble you…its not flirting. its trying to chip away at your self esteem lmao i have literally had men tell me “i’m not complimenting you because everyone else does that”)
Wtf
Tweet, 19:46 UTC · 5 likes
Hey I agree, but somehow I don’t think vandalizing birdwatching signs is part of a viable theory of change
Reply to @littIeramblings, 19:53 UTC · 2 likes
sarah (@littIeramblings): hello
in celebration of me recently surpassing 2k followers and having no plans on a Friday night, AMA
Do u have goss
Reply to @littIeramblings, 20:09 UTC · 9 likes
sarah (@littIeramblings): hello
in celebration of me recently surpassing 2k followers and having no plans on a Friday night, AMA
What’s the internal psychology of playing the “I’m just a girl idk much” character despite clearly being plenty knowledgeable+smart+capable?
Tbc I’m not saying this bad or dishonest (“character” isn’t really the perfect word to use here)
Tweet, 20:16 UTC · 65 likes
I think now would be a good time to pause AI scaling and ask all the cracked engineers to instead do the trivial, meaningless task of expanding and integrating current-level systems (and potentially new systems of similar scale) into all of society
Thread, continued, 20:19 UTC · 5 likes
Correct take
Thread, continued, 20:19 UTC · 6 likes
There’s a single Everettian branch where Ilya et al work on applying GPT-4 level compute to cultured meat and human health+wellbeing and I become optimistic about the future
Reply to @littIeramblings, 20:29 UTC · 4 likes
sarah (@littIeramblings): ‘clearly’
Saturday, September 14, 2024
Reply to @datepsych, 02:34 UTC · 6 likes
(The tweet being replied to by @datepsych is no longer available.)
I don’t like the term “positive illusions” then - this is just what you’d expect from heterogenous values + selection effects
Reply to @Duderichy, 02:35 UTC · 3 likes
the Rich (@Duderichy): who actually has balanced takes on ai rn
You want balanced or accurate?
Reply to @Duderichy, 02:36 UTC · 1 like
the Rich (@Duderichy): @ryxcommar Yes
Paul Christiano
Sunday, September 15, 2024
Tweet, 00:25 UTC · 18 likes
Winter but it’s light 16 hours/day
Quote tweet, 02:31 UTC · 32 likes
Reddit-like interface for Twitter!
If you want it to work for you, upload your data to the Archive!
Quoting ❤️🔥 xiq (@exgenesis): reminder to request your twitter archive :)
Thread, continued, 02:31 UTC · 2 likes
Bc thumbnails probably make people click
Reply to @theshahidhn, 02:54 UTC · 2 likes
shahid (@theshahidhn): can someone explain why is it that when I eat at a slight caloric deficit I seem to lose weight, but when I eat at a larger caloric deficit I seem to maintain or even put on weight??
No
Tweet, 07:20 UTC · 16 likes
When Claude starts the response with “interesting question!“
Quote tweet, 07:28 UTC · 17 likes
Being against rodent studies sounds like bleeding heart animal lover cope but they’re in fact shockingly bad (like truly much less helpful than you’d naively expect)
Quoting Someone:
Thread, continued, 07:28 UTC · 2 likes
Of course in principle if science and biomedical research were populated by ideal Bayesian agents, they could just be well-calibrated about how much evidence mice provide
But in reality everyone over-indexes and we cure cancer a gazillion times in mice and then not in humans
Thread, continued, 07:28 UTC · 2 likes
Good paper https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2642860/
Reply to @___frye, 17:32 UTC · 2 likes
frye (@___frye): FWIW my wife had this exact experience recently after testing positive for the MTHFR gene mutation
started taking methyl folate. chronic depression symptoms pretty much disappeared
it’s been more effective than SSRIs, which she tried for years
Deplin go brr
Reply to @elder_plinius, 20:35 UTC · 54 likes
Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭 (@elder_plinius): link to OP’s chat: https://chatgpt.com/share/66e718e5-c934-8001-8ffa-ef7ca1f165ff
Now we wait
Tweet, 23:25 UTC · 4 likes
Seems like plausibly a false flag account. I want to say it probably is, in fact. Just FYI
Monday, September 16, 2024
Quote tweet, 07:48 UTC · 13 likes
What the fuck
Quoting Sridhar Ramesh (@RadishHarmers): Bob wins with probability 528,495,731,290,486,310,231,292,772,358 / 2^100, while Alice wins with lesser probability 395,447,854,881,692,553,836,159,303,474 / 2^100.
Reply to @SpacedOutMatt, 14:39 UTC · 2 likes
Matt 🔸 (@SpacedOutMatt): @littIeramblings @AaronBergman18 @jskf__ @Laura_k_Duffy 👀👀👀
👀
Tweet, 15:57 UTC · 8 likes
Real friends share their data with u 🥰 @NoahTopper
Reply to @NoahTopper, 15:57 UTC · 2 likes
Quinoah 🔍⏸️ (@NoahTopper): @AaronBergman18 @ilex_ulmus Oh hey, I just downloaded my raw Ancestry DNA data: https://drive.google.com/file/d/1nzZ3W4qY-zmYB3cEGkfC1KPd5SqpnKRa/view?usp=sharing
Jinx
Tweet, 21:21 UTC · 77 likes
Moral anti-realist rationalists when they dislike whatever you’re doing
Tuesday, September 17, 2024
Reply to @Yozarian22, 01:44 UTC · 3 likes
Yozarian22 (@Yozarian22): @AaronBergman18 I don't see the connection with anti-realism?
Realism lets you appeal to the observer-independent badness of an action when trying to convince someone to change their behavior
My sense is that many/most anti-realists think that, at the end of the day, moral claims are just claims about someone’s preferences…
Thread, continued, 01:47 UTC · 3 likes
If I’m playing loud music late at night, a (consistent) anti-realist can ask me to stop but won’t claim that what I’m doing is wrong in some absolute sense; it’s just my preference vs theirs, and their goal is to change my preference to “not keep playing loud music rn”
Reply to @joejanizek, 11:27 UTC · 2 likes
Joseph D. Janizek (@joejanizek): Reasons I’m somewhat skeptical of these posts are: A) there’s already widespread folate supplementation in most processed foods, and B) there’s basically no downside to folate/B12 supplementation, so if you were worried about it, you could probably buy a lifetime supply for..
The whole point is that folate supplementation doesn’t work for some people. They need methylfolate
Reply to @norvid_studies, 11:28 UTC · 3 likes
norvid_studies (@norvid_studies): +5-10 iq intervention with no downsides that affects 1 in 5 people and not widely used? doesn't sound impossible but ... ?
(quoting @amplifiedamp: Get your MTHFR gene tested!!! Doing so and taking methyl folate boosted my IQ by 5-10 points permanently 20% of Americans have the mutation, get a GeneSight at-home test)
Depression causally reduces IQ (I think) and deplin = methylfolate is approved as an antidepressant so at least directionally there should be some genuine effect here
Tweet, 12:49 UTC · 9 likes
Moral value realism (state A of the world is objectively better than state B)
//
Normative realism (it’s objectively true that you should do X)
Thread, continued, 12:50 UTC · 7 likes
Cc @absurdlymax @SpacedOutMatt
We’re hiring a guy to be value anti-realism + normative realism to complete the grid
Apply here
Thread, continued, 12:51 UTC · 3 likes
Fr though it’s strange to me that these rarely get conceptually separated
Reply to @ShakeelHashim, 13:03 UTC · 3 likes
Shakeel (@ShakeelHashim): Claude is great, but one weird thing is that it's so sycophantic I don't really *respect* it?
You’re absolutely right! That is very helpful information.
Tweet, 13:15 UTC · 12 likes
While I don’t really want my whole feed to be the little hippo, I also can’t be expected to not like hippo posts when they appear
I’m not sure there’s a way out
Reply to @absurdlymax, 13:24 UTC · 2 likes
Max Alexander (@absurdlymax): @AaronBergman18 @SpacedOutMatt That's because half of these views are insane
☹️
Quote tweet, 13:50 UTC · 8 likes
One of many interesting things in the quoted graph is that placebo group weight returns to baseline
This isn’t what you’d expect from regression to the mean, which occurs with other systemic, semi-mysterious diseases like depression
Quoting Crémieux (@cremieuxrecueil): The worst common argument against Ozempic is that if you stop using it, the effect wears off.
Of course it does! If you stop exercising and dieting, you put on weight from that too, but that's not enough for anyone to seriously contend that diet and exercise are bad.
Quote tweet, 14:37 UTC · 4 likes
Turns out this is closer to literally true than I thought
Quoting Aaron Bergman 🔍 (@AaronBergman18): Aaron don’t wear his shirt inside out for a whole day difficulty impossible
Reply to @arithmoquine, 22:31 UTC · 3 likes
henry (@arithmoquine): this is probably largely why openai hides o1's COT.
i suspect:
- RL has made it increasingly alien- they don't want to implicitly incentivize it to act human. it should be a black box, only concerned with getting the correct answer
(quoting @arithmoquine: if your model's chain of thought is remotely legible, i don't believe in it. are you really telling me simulating a human's internal monologue is the optimal mode of reasoning? Compress Harder. i don't trust the efficiency of an intelligence that thinks anything like us)
I mean all its raw intelligence comes from predicting and understanding human text
I have a hard time thinking about how much that matters but intuitively it should strongly bias the model towards using in-distribution (ie human-legible) reasoning tokens
Tweet, 22:50 UTC · 9 likes
Did we ever find out why Elon killed circles?
It was the best feature of any social media app anywhere, it worked perfectly fine, and then poof
Wednesday, September 18, 2024
Reply to @jaybaxter, 01:05 UTC · 3 likes
Jay Baxter (@jaybaxter): @AaronBergman18 >It was the best feature of any social media app anywhere
This part is debatable :)
If you would please consult the graph
Thread, continued, 01:06 UTC · 3 likes
However I have requested a community note just in case
Reply to @jakobgreenfeld, 08:16 UTC · 2 likes
Jakob Greenfeld (@jakobgreenfeld): The biggest life hack is making sure you get your dopamine exclusively from activities that move you closer to your ideal life.
Darn
Quote tweet, 17:19 UTC · 30 likes
Holy shit lmao
That’s *digits* in each number
Next token prediction really does go brr
Quoting Smoke-away (@SmokeAwayyy): Multiplication using o1-mini vs GPT-4o
Reply to @RokoMijic, 17:30 UTC · 3 likes
Roko 🐉 (@RokoMijic): @AaronBergman18 this is still worse than human experts
heck, even I could probably beat this with a pen and paper (though obviously more slowly)
Yeah, but the point is that next token prediction ends up generalizing extremely well. Now we’re all used to this, but I wouldn’t have expected it
Reply to @dylanmatt, 18:30 UTC · 5 likes
(The tweet being replied to by @dylanmatt is no longer available.)
This would probably be true beneath some trivial amount
I’d rather be paid $0 than $1. Probably $10 even. But at realistic amounts you’re right
Quote tweet, 18:31 UTC · 17 likes
I don’t care what the correct plural form is currently. It should be octopi
Quoting Crémieux (@cremieuxrecueil): Amusingly, Democrats were more likely to get the correct answer ("Octopuses", not "Octopi").
Reply to @HenriThunberg, 18:40 UTC · 2 likes
Henri Thunberg 🔸 (@HenriThunberg): Boooh, rare Aaron L.
We shouldn't nitpick what people say, but changing what it "should be" is withering down language to a goo where no beauty can be found.
😢
Reply to @albrgr, 19:57 UTC · 4 likes
(The tweet being replied to by @albrgr is no longer available.)
Big if true
Tweet, 20:09 UTC · 4 likes
Strava has fallen
Thursday, September 19, 2024
Reply to @polisisti, 01:41 UTC · 2 likes
Alessandro (@polisisti): @AaronBergman18 explain your position
No
Tweet, 03:56 UTC · 31 likes
Society (metaphorically) if I could write as easily and quickly on Hinge and the EA Forum as I do on twitter
Thread, continued, 03:58 UTC · 6 likes
They’re two totally different modes
Low stakes: I can just type and my thoughts magically appear on the screen
High stakes: I can maybe write a sentence but then I’ll delete it bc it’s clearly not *exactly* what I want to say
Reply to @daniel_271828, 04:03 UTC · 3 likes
zzDANIEL_ETH (@daniel_271828): Me, writing a normal email to one person I know and am on good terms with: [let me triple read this over first to make sure there's no embarrassing typos]
Me, writing an edgy tweet for the entire word to see: [done and send tweet; close app]
Real
Reply to @chismosavirus, 17:46 UTC · 4 likes
megan (@chismosavirus): u can always tell when a man is dating someone new. why you going to aquarium and griffith observatory
Maybe I just like looking at fish
Friday, September 20, 2024
Tweet, 04:09 UTC · 7 likes
New Spotify Wrapped just dropped. Sorry for being a basic Type of Guy
Reply to @austinc3301, 04:29 UTC · 7 likes
Agus 🔸 (@austinc3301): @pli_cachete Fun fact: I once had one of these with MC Hammer, over whether the theory of evolution was right or not.
What in the world
Reply to @mattyglesias, 23:44 UTC · 3 likes
(The tweet being replied to by @mattyglesias is no longer available.)
Seems bad
Saturday, September 21, 2024
Reply to @RafaRuizdeLira, 00:11 UTC · 2 likes
Rafael Ruiz ⏸️🔸 (@RafaRuizdeLira): Would beat Moo Deng in a fight
(quoting @CincinnatiZoo: Fritz the baby hippo is two years old and now weighs 1,345 pounds! 👏)
They would never fight
Reply to @TechEmails, 01:32 UTC · 16 likes
Internal Tech Emails (@TechEmails): Mark Zuckerberg emails WhatsApp cofounder
April 6, 2013
Aww he’s shooting his playdate shot
(I unironically respect this)
Reply to @JesseTayRiver, 18:00 UTC · 2 likes
Jesse Smith (@JesseTayRiver): @GjMcGowan Washing machine at the new crib?
Tweet, 18:28 UTC · 7 likes
I had a dream about not being able to solve a leetcode problem at a Jane Street interview
Reply to @datepsych, 18:34 UTC · 7 likes
(The tweet being replied to by @datepsych is no longer available.)
It’s a good example of why “social injustice” isn’t an especially useful term
It is what it is - life isn’t fair, and there is suffering, and those are bad things that we can sometimes fix or partially fix
Sunday, September 22, 2024
Reply to @izzyz, 02:23 UTC · 3 likes
izzy (@izzyz): isn’t it weird to imagine that us humans and animals are the only living things in the universe? this easily seems like a reason everyone should believe aliens are real
Animals are actually pretty weird if u think about it
But yes I agree
Reply to @adrusi, 03:11 UTC · 9 likes
autumn (@adrusi): i had put some stock in the "dei killed the chips act" narrative because i heard it from ezra klein, who i trust not to be insane about this stuff and takes his work seriously
very happy to find it's just another way i underestimated the biden admin. fantastic news, bravo
(quoting @Noahpinion: Remember all those articles screaming that "DEI killed the CHIPS Act" and "Industrial policy has failed" and blah blah blah? Well, the TSMC Arizona factory is now up and running and making chips, and it's ahead of schedule. 🇺🇲🇺🇲🇺🇲)
Aside from any object level stuff, I will chime in to say I think this is basically the right way to form opinions on matters you don’t personally have a near-term, direct stake in 80% of the time
Thread, continued, 03:15 UTC · 8 likes
There are like 7 things I have a good inside view on, the rest I just defer to some weighted avg of Ezra Klein, Matt Yglesias, Brian Tomasik, Rob Wiblin, and Scott Alexander (this is slightly hyperbolic tbc)
Quote tweet, 12:16 UTC · 3 likes
You can’t convince me Stein exists in-between election cycles
Quoting PBS News (@NewsHour): The Supreme Court has rejected an emergency appeal from Nevada’s Green Party seeking to include presidential candidate Jill Stein on the ballot in the battleground state. https://to.pbs.org/3TxGLj6
Quote tweet, 12:34 UTC · 3 likes
Rip
Quoting Liam Bright (@lastpositivist): @besttrousers I don't think we propose to so degrowth we just do it.
Reply to @jskf__, 19:06 UTC · 3 likes
jskf 🔀 (@jskf__): @niplav_site Yes, I often use "I believe" to suggest much lower confidence.
Yea, people pretend not to know it’s a rhetorical tool
Same with “I think”
Reply to @jskf__, 19:10 UTC · 4 likes
jskf 🔀 (@jskf__): What do you mean by pretend to not know?
Maybe that wording is a little harsh, but I see people say that there’s no semantic difference between “X” and “I believe/think X”
Thread, continued, 19:11 UTC · 3 likes
Whereas I think they implicitly do understand the difference
Quote tweet, 19:49 UTC · 17 likes
Someone more conscientious than me should do this
Quoting Holly ⏸️ Elmore (@ilex_ulmus): I do NOT have time to run any more clubs, but I have this dream of a once a month “Get the Ball Rolling” meeting where we all co-work on important non-habit-based life improvements like setting up a Vanguard account, starting the Accutane process, signing up for Cryonics, etc.
Monday, September 23, 2024
Reply to @SpacedOutMatt, 02:15 UTC · 2 likes
Matt 🔸 (@SpacedOutMatt): I'm also getting a physical set printed and will be filming an intro and game with @AaronBergman18, @Laura_k_Duffy, and @arthur_wright_ when it arrives! And the physical set will be coming with me to EAG Boston in November!
👀big if true 👀
Reply to @matsonj, 06:31 UTC · 14 likes
Jacob Matson (@matsonj): Related: every app is just an interface for a database
(quoting @vboykis: Everything is information retrieval if you squint hard enough)
Not doodle jump
Reply to @metakuna, 12:26 UTC · 2 likes
metakuna (14/100 blog posts) (@metakuna): what did pigeons eat before crumbs were invented?
Other pigeons
Reply to @LuiseWoehlke, 12:32 UTC · 4 likes
Luise Woehlke (@LuiseWoehlke): The fraction of autistic people in EA is insane
😈
Reply to @EAForumPosts, 13:03 UTC · 5 likes
EA Forum Posts (@EAForumPosts): New popular post from the EA Forum:
"Announcing Animal Welfare vs Global Health Debate Week" by Toby Tremlett🔹
https://forum.effectivealtruism.org/posts/9ad4C4YknLM5fGG4v/announcing-animal-welfare-vs-global-health-debate-week
👀
Quote tweet, 13:14 UTC · 2 likes
👀
Quoting EA Forum Posts (@EAForumPosts): New popular post from the EA Forum:
"Announcing Animal Welfare vs Global Health Debate Week" by Toby Tremlett🔹
https://forum.effectivealtruism.org/posts/9ad4C4YknLM5fGG4v/announcing-animal-welfare-vs-global-health-debate-week
Quote tweet, 18:42 UTC · 20 likes
It really never gets old
The wings aren’t big enough to make the plane not fall down (intuitively; evidently somehow they do)
When I stick my arms out and run as fast as I can, I don’t even come *close* to starting to float above the ground
Quoting Aaron Bergman 🔍 (@AaronBergman18): Guy who prefers living near an airport bc the sound is pretty minor and the planes are cool to look at (me)
Thread, continued, 18:45 UTC · 2 likes
The surface area of the wings just isn’t that big. And commercial jets are >150k pounds. It just doesn’t add up
(video, plays on X)
Reply to @timfduffy, 18:53 UTC · 5 likes
Tim Duffy (@timfduffy): @AaronBergman18 You just need to run 100x faster and make your arms more wing-shaped, skill issue
Hmm ok thanks for the feedback. I’ll try this
Reply, 19:06 UTC · 1 like
I think there’s some empirical evidence for this at <1 mile. I’m ~4 miles away. Honestly I have no idea, someone should tell me if I should be worried. Intuitively it doesn’t feel like trying to time talking like that would make any difference
Reply to @LabGrownNeet, 19:09 UTC · 2 likes
laს გr🤗wn იeet (@LabGrownNeet): @AaronBergman18 Oh I mean I just literally had to bc of the noise
Ohh lol
Reply to @SpacedOutMatt, 20:38 UTC · 2 likes
Matt 🔸 (@SpacedOutMatt): If you were running at the takeoff speed of a 737 (150 mph) I think your intuitions would change
Why would it help to slow down?
Tweet, 21:32 UTC · 9 likes
The more I interact with the world the more correct this seems to be
Quote tweet, 21:52 UTC · 11 likes
Whenever someone says “there have been numerous waves of AI hype and disappointment”
Quoting bayes (@bayeslord):
Reply, 22:00 UTC
@SpacedOutMatt U ARE A LIAR
Reply to @SpacedOutMatt, 22:39 UTC · 2 likes
Matt 🔸 (@SpacedOutMatt): Given that you were literally a math major this is not an excuse :P
Still fake imo
Reply to @SpacedOutMatt, 22:48 UTC · 3 likes
Matt 🔸 (@SpacedOutMatt): If u spent as much time with an aerodynamics textbook as u did tweeting about how you don't understand planes, maybe you'd understand planes 😈
Help I’ve been murdered 💀💀💀
Reply to @SpacedOutMatt, 22:49 UTC · 3 likes
Matt 🔸 (@SpacedOutMatt): @AaronBergman18 @ethai__ https://x.com/SpacedOutMatt/status/1810634561069465924 I have literally taken him to the planes museum and explained how the planes work
We must retvrn
Tuesday, September 24, 2024
Reply to @JesseTayRiver, 02:47 UTC · 2 likes
Jesse Smith (@JesseTayRiver): @SpacedOutMatt @AaronBergman18 @ethai__ Both of you have made some really good points in this discussion. I don't know who to trust!
Me imo. Matt works in airplanes so he’s biased and can’t be trusted
Reply to @RichardMCNgo, 11:31 UTC · 5 likes
Richard Ngo (@RichardMCNgo): “If [moral] nihilism is true, nothing matters - there’s not reason to do one thing over another. So in our deliberation, we can act as if [moral] realism is true.”
I’ll unironically defend this FWIW
Reply to @herbivoryze, 12:41 UTC · 10 likes
HerbivoRyzing (@herbivoryze): The first academic paper specifically on turning predators into herbivores
Based
Reply to @manic_pixie_agi, 12:56 UTC · 2 likes
Corinne (🏠 in NYC) (@manic_pixie_agi): @AaronBergman18 this sounds made up. 100%? Maybe a skill issue but I can't find the original paper from a google search
Nooo nothing on the internet is false noo
Reply to @RichDecibels, 14:29 UTC · 3 likes
Richard D. Bartlett (@RichDecibels): you have not found smarter people to be less kind, on average?
I don’t think so, no, but regardless they *needn’t* trade off
Tweet, 17:33 UTC · 7 likes
There is exactly one functional part of US healthcare and it is that somehow it’s actually quite easy and totally free to get COVID+flu vaccine
Quote tweet, 18:37 UTC · 15 likes
Is this true? If so I feel like it’s crazy under-discussed
Same as with SBF - ideological, material, and psychological explanations for behavior are hardly sufficient when the person in question is permanently hypomanic
Quoting Nemo (@thecaptain_nemo): hitler was the bryan johnson of his day but his doctor was basically just winging it
check out hitler’s supplement stack. would receive up to 20 injections a day. 70+ different substances including testosterone, adrenaline and other hormones, methamphetamine, cocaine, heroin...
Reply to @jkeatn, 18:39 UTC · 4 likes
Jake Eaton (@jkeatn): This is kind of the entire German army though
They got amphetamines, probably not the whole storm cocktail (right?)
Tweet, 20:32 UTC · 17 likes
Try to list the 10 most populous countries
How many of the ten did you get right, if order doesn’t matter?
Correct list below
Thread, continued, 20:32 UTC · 4 likes
Thread, continued, 20:32 UTC · 3 likes
I’m surprised I did this well
Reply to @timfduffy, 21:00 UTC · 6 likes
Tim Duffy (@timfduffy): @AaronBergman18 Felt good about my 9 until seeing that's the median, you have knowledgeable followers
Yeah this is extremely typical smh 😒
Reply to @littIeramblings, 21:09 UTC · 12 likes
sarah (@littIeramblings): I'm looking for work! I recently decided that I was motivated enough to help reduce AI risk to quit my full-time job & pursue opportunities in AI safety comms/governance.
I'm currently undertaking a career transition period funded by the Long Term Future Fund (more below)
Based, twitter go brr
Reply to @tenobrus, 21:17 UTC · 3 likes
Tenobrus (@tenobrus): were u pretending to be an adult online since u were a small child?
I’ve been 16 since I was 12
Quote tweet, 22:19 UTC · 7 likes
Warning in case anyone thinks I’m smart
Quoting techsiren ♥︎ (@techhsiren): sometimes u think a man is smart but he's really just regurgitating some shit he heard on a podcast word for word
Reply to @WillowChem, 22:42 UTC · 2 likes
MagpAI (@WillowChem): Seriously how the fuck does anyone get addicted to amphetamines they suck so bad
For better or worse IME they are the opposite of this for me (every dimension gets better, only downside is tolerance)
Reply, 22:43 UTC · 1 like
😢
Wednesday, September 25, 2024
Tweet, 03:03 UTC · 2 likes
Kinda wild that plausibly the most consequential thing ✨wokism✨ will have achieved causally runs through Biden promising to pick a black woman as VP
Not saying that this is good or bad tbc, but POTUS matters quite a bit
Reply to @Mjreard, 04:48 UTC · 2 likes
Matt Reardon (@Mjreard): Am I wrong to think Chicago, London, and the Bay are better (based mostly on comparing my experience to what people in DC say about DC)?
My sense is that the bay is better on people just saying what they think out loud, but this isn’t identical to or sufficient for being “earnest, thoughtful, and great”
Quote tweet, 18:42 UTC · 83 likes
1) what
Quoting Ivanka Trump (@IvankaTrump): Leopold Aschenbrenner’s SITUATIONAL AWARENESS predicts we are on course for Artificial General Intelligence (AGI) by 2027, followed by superintelligence shortly thereafter, posing transformative opportunities and risks.
This is an excellent and important read :
https://situational-awareness.ai/
@leopoldasch
Reply to @Bbburner19, 18:57 UTC · 2 likes
The People Mover That Was Promised (@Bbburner19): @AaronBergman18 She should endorse Kamala.
I agree
Reply to @krishnanrohit, 19:51 UTC · 2 likes
rohit (@krishnanrohit): What does this mean?
- Sam writes blogpost about being 1000s of days from superintelligence
- Mira jumps ship
- Ivanka tweets Leopold(quoting @miramurati: I shared the following note with the OpenAI team today.)
Probably not directly related
Thursday, September 26, 2024
Quote tweet, 00:05 UTC · 7 likes
Seems illegal
Quoting Mario Nawfal (@MarioNawfal): 🚨 BREAKING: OPENAI TO REMOVE NON-PROFIT CONTROL, GIVING SAM ALTMAN $150 BILLION EQUITY
OpenAI is planning to restructure its business into a for-profit benefit corporation, removing control from its non-profit board.
This shift, aimed at attracting more investors, will also grant CEO Sam Altman equity for the first time, potentially valuing the company at $150 billion.
The non-profit will retain a minority stake, while concerns arise about AI safety as governance changes are finalized.
Source: Reuters
Tweet, 00:53 UTC · 10 likes
I have to know if @QualyThe says their own name like this, like a Pokemon
(video, plays on X)
Quote tweet, 01:04 UTC · 4 likes
.@SpacedOutMatt: in 7 minutes I’m gonna do a mock job interview with Aaron. I bet he’s studying hard and taking notes
Me:
Quoting Aaron Bergman 🔍 (@AaronBergman18): I have to know if @QualyThe says their own name like this, like a Pokemon
Reply to @austinc3301, 02:25 UTC · 5 likes
Agus 🔸 (@austinc3301): @RomanHauksson @ciphergoth @AaronBergman18 🤝
What
Quote tweet, 02:25 UTC · 12 likes
.@OpenAI do u need a guy
Quoting Aaron Bergman 🔍 (@AaronBergman18): @slimepriestess You can always make terrible models for fun (this took 12 hours on an m1 MBP)
Reply to @austinc3301, 02:38 UTC · 3 likes
Agus 🔸 (@austinc3301): I hear they’re looking for a new Chief Research Officer
@sama I’ll plan for you to pick me up in that fancy car. Meet at the white house tomorrow at noon
Reply to @QualyThe, 03:49 UTC · 6 likes
Qualy the lightbulb (@QualyThe): Real
Qualy answer me so I can rest my weary brain
Reply to @elonmusk, 03:50 UTC · 5 likes
Elon Musk (@elonmusk): This is deeply wrong
(quoting @ParikPatelCFA: How to make $10 billion dollars: - Raise $50 million from Elon Musk to start a nonprofit - Tell everyone you are doing this for the sake of humanity and raise billions - Convert from non-profit to for-profit and grant yourself equity)
You should sue
Tweet, 03:51 UTC · 65 likes
OpenPhil gets $50B lawsuit payout arc?
Thread, continued, 03:56 UTC · 22 likes
Fr they should sue. With Musk and whoever else
Reply to @freed_dfilan, 04:04 UTC · 25 likes
Daniel Filan 🔎 (@freed_dfilan): @AaronBergman18 ?
Should sue OpenAI for taking their money as a nonprofit and then going for-profit
Reply to @besttrousers, 04:13 UTC · 9 likes
Matt Darling 🌐🏗️ (@besttrousers): OpenAI has 1700 employees according to Wikipedia.
If their quit rate is the same as the economy-wide average, you would expect 36 people to quit a month, or ~1.7 a day.
This is good context, but what’s the churn rate for jobs that pay a gajillion dollars per millisecond
Reply to @TylerAlterman, 04:31 UTC · 2 likes
Tyler is finishing a book, slow to reply (@TylerAlterman): What best characterizes your feeling about OpenAI
3rd option but also great talent pipeline for anthropic
Quote tweet, 04:34 UTC · 74 likes
Ideal woman:
- female
- likes me
Quoting mikayla (@honeyNonABG): My ideal man:
- 6’0+
- Good facial bone structure
- 120+ IQ
- Deep voice
- Funny
- Assertive
- Cowboy personality
- Income > $130K(Optional)
- Ivy League educated
- Old money
Quote tweet, 12:48 UTC · 7 likes
👇
Quoting Aaron Bergman 🔍 (@AaronBergman18): Yeah these recent revelations have pushed me over the line from “way too self confident but earnest” to “basically a bad person who can’t be trusted”
Reply to @beala, 15:24 UTC · 2 likes
Alex Beal 🆎 (@beala): Oh yikes.
No this is based. I had a poster of a duck up at one point bc I thought it was funny but then I moved and it disappeared
Reply to @TylerAlterman, 15:32 UTC · 2 likes
Tyler is finishing a book, slow to reply (@TylerAlterman): More from her:
"Courting women is and should be risky. Men are supposed to take well-calibrated risks, that's a huge part of masculinity"
Reply to @francip, 21:40 UTC · 2 likes
Franci Penov (@francip): @AaronBergman18 You get only one of the two.
(gif, plays on X)
Reply to his own earlier tweet, 21:55 UTC · 5 likes
Aaron Bergman, September 26, 2024: Fr they should sue. With Musk and whoever else
Manifold market https://manifold.markets/AaronBergman18/will-openphil-or-a-related-entity-e
Thread, continued, 21:58 UTC · 4 likes
I also hope this happens for the purely selfish reason that it’ll be incredible to watch/follow
Reply to @lxrjl, 21:59 UTC · 2 likes
alex lawsen (@lxrjl): How many people can an organisation lose, hypothetically, before it should start worrying that maybe it's getting close to being "without its people"?
Well if you replace one employee with an identical counterpart, and then another, and then another…
Reply to @animalologist, 22:02 UTC · 4 likes
taco belle (@animalologist): i have had multiple men describe me as "psychoactive" and "addictive" like buddy i am not a drug i'm just a girl
(quoting @sighswoon: Can’t tell you how many Men have accused me of doing witchcraft on them (never did) instead of being responsible for the intense passionate feelings that arose inside them)
Friday, September 27, 2024
Reply to @NathanL0lz, 09:48 UTC
Nathan (נתן) 🌹 ⬱ ✡︎ ⚣🌂❌❌❌🍉 (@NathanL0lz): Confused as to how python is partially responsible for the end of the world
https://80000hours.org/problem-profiles/artificial-intelligence/
Reply to @NathanL0lz, 09:57 UTC · 5 likes
Nathan (נתן) 🌹 ⬱ ✡︎ ⚣🌂❌❌❌🍉 (@NathanL0lz): I don’t see how it’s python specifically that’s responsible for this?
Most of the progress in AI has been made using python (libraries like PyTorch and tensorflow)
But it’s kinda a joke - there’s nothing super special about python that enables modern AI, it’s just how things have gone
Tweet, 11:21 UTC · 5 likes
There’s probably an economic case for broader Medicaid eligibility purely via reducing the personal risk of entrepreneurship relative to employment
Tweet, 12:46 UTC · 13 likes
In hindsight it’s kinda weird I was put in both a social skills class and some sort of sensory therapy (idk what it’s formally known as, they gave me one of these brushes) and yet I don’t remember anyone saying the word “autism” until a few months ago
Thread, continued, 12:46 UTC · 5 likes
I don’t think this was bad tbc. Tentatively it seems good, actually. I’m not sure the label would have helped me. But still interesting
Reply, 13:40 UTC · 6 likes
Good question. Probably the transatlantic slave trade: Claude estimates that it resulted in something like a billion life-years lived as slaves. Hard to think of anything worse for humans
Tweet, 13:41 UTC · 3 likes
Good question, and one I hadn’t thought about. My best guess below
Thread, continued, 13:41 UTC · 4 likes
Reply to his own earlier tweet, 13:43 UTC · 3 likes
Aaron Bergman, September 27, 2024: Good question. Probably the transatlantic slave trade: Claude estimates that it resulted in something like a billion life-years lived as slaves. Hard to think…
Misread - about half a billion actually
Reply to @ptrschmdtnlsn, 14:19 UTC · 4 likes
Peter Schmidt-Nielsen (@ptrschmdtnlsn): When you watch videos of fabs or even just PCB shops, jeez, the fact that a microprocessor costs less than $100k is a miracle.
My monkey brain refuses to believe computers are real at all
Tweet, 14:41 UTC · 14 likes
(Valiant formatting effort by Claude)
Tweet, 20:17 UTC · 31 likes
Bro my lightbulbs are trying to sell me an AI assistant
Thread, continued, 20:17 UTC · 4 likes
Cc @QualyThe
Reply to @QualyThe, 23:59 UTC · 2 likes
Qualy the lightbulb (@QualyThe): I would never do that bb
❤️
Saturday, September 28, 2024
Reply to @polisisti, 13:44 UTC · 1 like
Alessandro (@polisisti): @Estimatrix "oschumus" simile ac "posthumous" anglice dicitur, sine littera P
Quid agitis, conservi Latine loquentes?
Reply to @polisisti, 13:51 UTC · 1 like
Alessandro (@polisisti): @AaronBergman18 @Estimatrix does ChatGPT know Latin 👀
Yes lol
Tweet, 14:29 UTC · 18 likes
Ok obviously I don’t think airplanes should be able to fly but also it’s crazy that we can get them to *land* in a specific spot
Like ex ante this seems ~impossible. It doesn’t seem like the kind of thing amenable to that degree of control
Thread, continued, 14:30 UTC · 3 likes
*gets blocked by @SpacedOutMatt*
Reply to @br___ian, 14:47 UTC · 5 likes
bri (@br___ian): the divide in flight understanders forever has been between guys who try to think about flight from first principles and guys who look at a bird and say "you're not so smart, how do you do it?"
How do they do it tho? Wild
Reply to @gbrl_dick, 15:03 UTC · 5 likes
Gabriel (@gbrl_dick): crossing the bering land bridge to the americas is the dumbest idea humans have ever conjured & i become irrationally annoyed anytime it gets brought up
like sure, let’s spend 6 months walking to get to the land beyond the sun god? probably covered in ice lol
(quoting @Villgecrazylady: Going to Mars is the dumbest idea humans ever conjured & I become irrationally annoyed anytime it gets brought up. Like sure thing, let’s spend a couple trillion $ to land on the next planet over where it’s -85°F all the time and there’s no oxygen or gravity or water. Brilliant!)
This but unironically
Reply to @shinboson, 15:39 UTC · 5 likes
𝞍 Shin Megami Boson 𝞍 (@shinboson): @AaronBergman18 The thing that made it click for me was that fluid “wants” to stick to surfaces and lift can be thought of as the counterforce to wedging an airstream away from a surface.
Yeah I’m too dumb for this intuition lol
Reply to @NoahTopper, 15:55 UTC · 3 likes
Quinoah 🔍⏸️ (@NoahTopper): @AaronBergman18 So I don’t disagree with you about planes, but I also don’t understand how anything else in the world works, so I see no reason to pick on planes in particular.
Yeah fr, but planes seem especially visceral to me bc like huge flying tube idk
Everything else parses as “magical trinket”
Reply, 16:21 UTC · 1 like
The philosophy I’ve read on this leads me to think it’s a totally vacuous question. It’s a word pointing to an empty/null concept
Reply to his own earlier tweet, 17:08 UTC · 2 likes
Aaron Bergman, September 28, 2024: Yeah fr, but planes seem especially visceral to me bc like huge flying tube idk Everything else parses as “magical trinket”
Wait actually there’s a more substantive difference: airplane flight doesn’t seem possible at all, totally apart from the question of how it works internally
Reply, 17:22 UTC · 1 like
Idk but I have the same issue. Lots of thoughts/takes, slow writer
Reply to @NoahTopper, 19:33 UTC · 7 likes
Quinoah 🔍⏸️ (@NoahTopper): Them: Hey do you wanna do [thing that makes you feel like shit]?
Me: No, that makes me feel like shit.
Them: Oh, come on, it's a special occasion! Live a little, let yourself do [thing that makes you feel like shit]!
Real
Sunday, September 29, 2024
Reply to @inesferhumi, 12:57 UTC · 2 likes
Inés (@inesferhumi): @AaronBergman18 Omg i was given one of those weird brushes too
Lol one of us🤝
Tweet, 13:15 UTC · 9 likes
For drugs with a long half life that take several weeks to reach steady state, why not just front load the dosing to reach target plasma levels quicker?
Dw I’m not chugging my bottle of levothyroxine, this is not a rhetorical question
Cc @FrostedSapling
Tweet, 14:50 UTC · 4 likes
Fair number of women on Hinge say something like “let’s be friends first and see where things go”
I’m guessing this makes things much harder for them—they too have been let down by society’s decision to corral dating into a distinct, digital-only sphere
Tweet, 15:00 UTC · 5 likes
Analytic philosophy needs to be a subsection in every library/bookstore/audiobook app
No lol I don’t want to read Meditations
Quote tweet, 15:02 UTC · 9 likes
New cause area: this but don’t fuck the mice up first, just give them massages
Quoting Grace (@kindgracekind): Me reading this headline: "Aw, they gave massages to mice! Mouse spa day!"
..."I wonder where they found all those injured mice, though"
...
☹️
Reply to @QiaochuYuan, 15:05 UTC · 2 likes
QC (@QiaochuYuan): i think it has but it's hard to tell. i especially need it when the air gets smoky from fires and the rest of the time it basically doesn't cost me anything to keep running it and i appreciate knowing it's there
+1
Reply to @appleMechanic, 15:10 UTC · 2 likes
teeming-asymptoting (@appleMechanic): @AaronBergman18 @QiaochuYuan or more re when fire season?
I mean I can’t know the counterfactual and the effect isn’t so strong and apparent that I can trivially detect it
But I’ve definitely noticed that I feel better when there’s fresh, clean air in general (ie outside in nature vs in a dusty room) and trust the filter mechanism
Reply to @QualyThe, 15:25 UTC · 2 likes
Qualy the lightbulb (@QualyThe): EA mfers be getting their politics from centuries old philosophy, meanwhile the next election will be decided by a 19yo using AI to churn out deepfakes of The Daily Show. bro you're worried about the wrong John Stewart mill
Banger
Tweet, 18:26 UTC · 3 likes
Kamala should run on making ADHD drugs easier to get. This is probably close to a bipartisan issue. Not even radical deregulation, just like let nurse practitioners prescribe and reduce the legal restrictions
Quote tweet, 22:38 UTC · 5 likes
Based
Quoting Cogent Sins🔎 (@Insect_Song): @eigenrobot I ❤️ people willing to die on hills.
Increasingly coming to think that a lot of society depends on disagreeable mildly autistic people.
Tweet, 23:08 UTC · 24 likes
Reply to @absurdlymax, 23:09 UTC · 3 likes
Max Alexander (@absurdlymax): Real
(quoting @austinc3301: sigh)
Real
Tweet, 23:55 UTC · 13 likes
Spotify clearly thinks I’m extremely cool
Monday, September 30, 2024
Reply, 03:02 UTC · 1 like
(The tweet being replied to is no longer available.)
Protestant-Jewish split household
Quote tweet, 11:50 UTC · 6 likes
Fine I’ll bite. What is Alexander Technique?
Quoting yatharth ༺༒༻ (@AskYatharth): woahh
i just had my first alexander technique session with olivia lin and
i didn't expect to feel this great. i feel tall, like i'm one body, i feel the floor separate
normally i feel like a potato chip bag trying desperately not to be here and my body feels all disjoint
Reply to @absurdlymax, 13:17 UTC · 2 likes
Max Alexander (@absurdlymax): @AaronBergman18
Goss 👀
Reply to @livgorton, 16:00 UTC · 1 like
Liv (@livgorton): i almost tagged aaron lol ✨
👀
Quote tweet, 17:15 UTC · 7 likes
Not to pick on this guy in particular but this meme doesn’t make sense. It should probably look something like this (random points with a slight bias towards the “true” ideology)
Quoting Zarathustra (@zarathustra5150): ✨
Reply to @HenriThunberg, 20:31 UTC · 7 likes
Henri Thunberg 🔸 (@HenriThunberg): The forum team has gone too far on this one
If you click it anyway it downloads and runs an exe that destroys your computer a la Stuxnet
Reply to @HenriThunberg, 20:52 UTC · 2 likes
Henri Thunberg 🔸 (@HenriThunberg): Locks your computer into a fullscreen google doc that you can only exit once you publish "that draft you've been working on"
I’d rather be Stuxnet’d
End of September 2024. Previous month: August 2024. Next month: October 2024. Back to the archive index.