leogao

Karma: 9,687

leogao 7 Mar 2026 6:32 UTC
1 point
0
in reply to: weberr13’s comment on: leogao’s Shortform
I’ve worked completely nocturnally before. it wasn’t the best experience in the world, and probably wouldn’t have been sustainable in the long run, but there are a lot of jobs out there that are way more demanding (submarine, space, oil rig).

leogao 7 Mar 2026 0:34 UTC
27 points
3
on: leogao’s Shortform
I’m very confused why purchasing power varies so dramatically internationally. like why are there countries where everyone has very low wages but everything is also really cheap so it balances out? prima facie, huge disparities like this should get evened out by arbitrage.
the simple explanation is that some labor can only be performed locally, labor mobility is limited (immigration laws, people don’t like moving, etc), and transportation costs for goods exist (shipping and tariffs).
however, global shipping is ridiculously cheap. and the economy increasingly consists of white collar jobs which could in theory be done remotely. for example, it seems it mind boggling to me that a top tier SWE/RS in the bay area is worth 10-100x more than one in India or Vietnam. like sure, someone being in the same timezone is great, and Zoom sucks, and so on. but for that price delta surely you could pay people to live nocturnally, construct apartments with bright lights synced to Pacific Time, invest in much better video call technology like that Google Beam thingy, etc?
maybe one possibility is that labor mobility is not actually that low for the very toppest tier people, and so if someone is actually worth that much then getting them good immigration lawyers is a trivial cost relative to their value/cost. so the market for the best people is very efficient and they all brain drain to the bay area the first chance they have. and so the lower salary of outsourcing is partly illusory at the highest levels, because the distribution of talent is very different across different localities, due to agglomeration effects.

leogao 5 Mar 2026 2:24 UTC
1 point
0
in reply to: Eli Tyre’s comment on: leogao’s Shortform
it feels rude to talk about specifics about other people. at a broad level, there are some people I’ve gained a lot of respect for. it’s easy for people to say they care about safety, so I don’t weigh that very heavily. but now I know who’s willing to step up and take actions in a crisis. and conversely too.

leogao 5 Mar 2026 2:14 UTC
6 points
3
in reply to: OVERmind’s comment on: leogao’s Shortform
it’s obviously about the department of war situation. it wasn’t intended to be vague, it just didn’t occur to me that it wouldn’t be obvious what it was about.

leogao 4 Mar 2026 5:34 UTC
7 points
1
in reply to: Thomas Kwa’s comment on: leogao’s Shortform
I think this comment aged very poorly

leogao 4 Mar 2026 5:33 UTC
14 points
6
on: leogao’s Shortform
I’ve learned a lot about various people around me, and myself, over the past few days.

leogao 28 Feb 2026 20:36 UTC
113 points
90
on: leogao’s Shortform
I claim that even if the openai contract is not meaningfully weaker safety wise, it is still bad for openai to publicly signal solidarity with ant but then sign with DoW.
suppose hypothetically the only difference between the openai and anthropic contracts is that the DoW wanted a snicker bar, and anthropic didn’t want to give DoW the snickers bar. even then, it would be a huge dick move for openai to publicly signal solidarity, and then sign with DoW to give them the snickers bar.

leogao 28 Feb 2026 10:33 UTC
9 points
0
on: leogao’s Shortform
the concept of a spontaneous unscheduled phone call is so strange and alien to me. you’re telling me there are people out there who want to be interrupted at random points in their day, and a large fraction of the time they are able to just pick up and talk? rather than constantly getting voicemails, and then leaving voicemails back because by the time you get around to replying, the caller is busy? do these people spend most of their days doing neither deep work nor being in social situations that would be rude to suddenly step away from?

leogao 27 Feb 2026 2:22 UTC
5 points
2
in reply to: Linch’s comment on: Linch’s Shortform
maybe i spend too much time inside the company and not enough time inside the set of people with the same astrological sign, but it feels to me like companies are often defined by a “vibe”, and that vibe happens to have been selected for profit maximizingness in the past (because if that vibe always sucks at profit it will go bankrupt), and companies will try to correct the vibe if they are obviously failing at maximizing profit, but most day to day decisions are guided by some mix of that vibe and a bunch of subagents that do random shit, and course correcting is really really fucking hard. i can easily think of half a dozen examples of companies being poorly described as primarily profit-maximizing with some minor divergences. google is a fountain of money with an AGI lab attached, and also a bunch of random side projects bolted on because doing a side project is good for your promo packet. meta is zuck’s personal kingdom and reflects his taste in exciting new directions. quant firms are basically statistical analysis firms that thumb their noses at deep learning. kodak was a film company that forgot to notice this was no longer a profitable vibe. ibm is a mainframe company that forgot to pivot into newer kinds of computers and is now desperately trying to pretend to be relevant in AI. tsmc is an ultra conservative fab company that has seen countless fads rise and fall and is scared to make big bets.

leogao 27 Feb 2026 2:10 UTC
6 points
0
in reply to: testingthewaters’s comment on: Thane Ruthenis’s Shortform
why is it not positive sum if you never slow down the model release schedule? positive sum doesn’t mean “nobody compromises on anything.” the entire point of positive sum trades is you find things that are compromises for the other party but ultimately not very costly for them, but are very beneficial to you; and in exchange, you do something that you would rather not do, but is not that costly for you, and beneficial for the other party. and on net both people get something they would rather have than the thing they sacrificed. so the potential trade here is:
- capabilities teams sacrifice x% of their compute and $y of salary money. they grumble a bit because on the margin another x% more compute could have make them very slightly faster, but at the end of the day it’s not make or break for them.
- alignment teams hire lots of really good alignment researchers, and do a huge amount of good work. alignment teams allow the lab to get some good PR. maybe even let their alignment work have the side effect of making the models very slightly better in ways that don’t reduce serial time a lot (though i think this requires more care than the PR).
  - of course this is all assuming that you actually do good work with the resources, which is pretty hard. if you’re doing useless work then it doesn’t matter whether you’re spending philanthropic money or lab money or whatever, you should stop doing bad work and do good work instead.
the superalignment situation is very unfortunate and reflects poorly on openai, but i think the entire situation was also net negative from a perfectly spherical openai’s perspective (negative PR from breaking this commitment far outweighed any of the benefits from making it in the first place; a perfectly spherical rational openai should have either not made the commitment at all, made a smaller commitment in the beginning, or upheld the commitment. in reality, the shift in position is of course easily explained by the board situation.)

leogao 27 Feb 2026 1:52 UTC
4 points
0
in reply to: Bronson Schoen’s comment on: Thane Ruthenis’s Shortform
no, i’m saying serial time is more of a bottleneck for capabilities than raw resources. diverting 5% of parallel resources is much less costly for capabilities than slowing serial speed by 5%

leogao 26 Feb 2026 22:59 UTC
12 points
2
in reply to: Thane Ruthenis’s comment on: Thane Ruthenis’s Shortform
i mostly agree that large sections of work at labs falls into traps somewhat like this. however, i have some key disagreements that make me more optimistic about it being possible to do (very) good work at a lab:
- it is possible to simply disagree with the lab’s actions in general, not to find any galaxy brain justification for why everything is well meaning. if you’re so principled that you are incapable of even existing inside any system which is somewhat misaligned, you’re going to have a rough time finding anywhere in this world that makes you happy, other than an isolated cabin in the woods.
  - i guess empirically some people struggle with disagreeing with the lab’s actions while at the lab. i agree these people should just not work at a lab. i think a prerequisite to someone doing good safety work at a lab is they should have a strong internal compass of what they believe is correct, and not be easily swayed by social pressure.
- the openai exoduses are a real thing, but it’s useful to view them in the context of an industry where it’s pretty rare for someone to be at the same place for many years, and a company which has had a higher turnover rate than average just in general, not only for safety people.
- i think you’re broadly right that safety stuff can’t cut too much into the capabilities stuff, but i contend that the main resource here is serial time rather than raw resources. this is important because it opens up some positive sum trades: capabilities people really hate it when their release dates get set back by e.g safety reviews (which is unfortunate), but they mind a lot less if you’re taking 5% of compute (which is a huge amount of absolute compute, and you should value a lot!), and they mind even less if you have an entire team doing good work somewhere out of their way (which you should also value a lot!).
  - i think it’s unproductive to think of the relation with labs as purely zero sum. there are areas where you can get some of what you value at low cost to them, and they can get some of what they want at low cost to you.
  - relatedly, imo the “good alignment work is good PR which is bad” argument is also classic zero sum thinking. good technical alignment work is not the maximally efficient PR investment; that would be using AI to help cute puppies. or to cure cancer. or to help cure cancer in cute puppies. you obviously don’t want to work on that stuff. but if you did actually good alignment work, and the company got some good PR out of it, is that the end of the world? in an extreme limit case, suppose openai solved interpretability tomorrow (or if you think interp is just capabilities, pick whatever field you like), like probably this is very net good, and you wouldn’t mind openai having a tiny sliver more goodwill in exchange.
- you, hypothetical lab employee, have no obligation to say untrue things in public to improve PR of your company. just say what you truly believe, within reason, and people will correctly model your beliefs.
  - maybe part of why this works for me is i truly do respect capabilities people a lot even if i disagree strongly with them about what is good for the world. i’m honestly quite similar to lab capabilities people in a bunch of ways, and i’m probably a lot more sympathetic to the position that maybe i’m wrong about all of this alignment stuff and that capabilities people are just being mostly reasonable than the average person on LW (though note that this is a statement about whether i can emotionally imagine being persuaded given the right facts; i am forbidden from using the emotional imaginability as an excuse to work on capabilities, and in fact i have not)

leogao 26 Feb 2026 9:38 UTC
0 points
1
on: leogao’s Shortform
the “move to ios” app is the worst piece of software i have ever used in my entire life

leogao 26 Feb 2026 7:03 UTC
6 points
0
on: leogao’s Shortform
sodium cotransport is really cool. while the gut can absorb glucose and sodium individually through several different pathways, there is a really important transporter (SGLT1) which carries glucose and sodium at the same time.
this is really important for rehydration. suppose you have cholera and vomit a lot and get super dehydrated as a result. drinking just water sucks, because you need to replenish the electrolytes that you’re losing too. but water with salts is still not optimal, because it’s absorbed less efficiently (also i think cholera interferes with some of the other absorbtion paths?). then one day some guy discovered cotransport, and invented a rehydration solution with not just salts but also glucose. this simple discovery and resulting invention has saved millions of lives
https://en.wikipedia.org/wiki/Oral_rehydration_therapy

leogao 26 Feb 2026 6:06 UTC
8 points
4
on: leogao’s Shortform
day 1 of using a new phone: there cannot be a single small bubble under my screen protector. it must be perfect.
day 1000 of using the phone: the square inch sized chunk of dead pixels on the screen is fine because it doesn’t usually cover anything important, and I can still read words in between the cracks

leogao 24 Feb 2026 21:00 UTC
3 points
0
on: leogao’s Shortform
TIL that Memphis, Tennessee is named after the ancient Egyptian capital Memphis

leogao 24 Feb 2026 20:01 UTC
4 points
0
in reply to: Sam Marks’s comment on: The persona selection model
first, I strongly disagree about the egregiousness and maliciousness of the examples. every day, many uneducated developers write code that is susceptible to SQL injections. this is especially true if your dataset is scraped from the entirety of github and therefore contains lots of older code, or code that was only ever intended for personal use and never meant to be secure. (using secure interpolation is kind of annoying and many people will just do the lazy thing!) the chmod example is less justifiable but I definitely know that people sometimes don’t want to deal with permissions and chmod 777 everything to make errors go away (this is common enough that some software intentionally forces you to use lesser permissions for important secrets! if everyone always followed the principle of minimum permissions this wouldn’t be necessary).
I still think you could have interpreted this as evidence for PSM if it had come out differently. suppose buggy code is very mildly correlated with maliciousness. obviously insecure code is more likely to be malicious, I don’t disagree with that. but if there are armies of shitty engineers out there and only a small handful of competent malicious adversaries, then p(malicious|insecure) is still very small, because of base rates. so observing that a model trained on insecure code is still non malicious 99% of the time would be evidence for PSM. whereas observing EM would be evidence against PSM, because it would imply p(malicious|insecure) close to 1, which is implausible given how many shitty engineers there are out there.
I think your argument here is still strongly tainted by hindsight bias, even if the first moment you saw EM you updated towards PSM. in my alternate mirror universe, the moment you saw not-EM you would have updated towards PSM. I think the thing that would prove non-hindsight requires you to pre-register whether you think EM is evidence for PSM before knowing whether EM is true.
I agree that the bird example is harder to spin. my argument would require there to be a nontrivial fraction of modern people to enjoy using outdated names for fun. since the Internet is vastly vastly bigger than all data from the 1800s, there are probably many incredibly weird and obscure Internet subcultures that would probably outweigh all data from 1800. however, even then, I have to admit “using archaic bird names” is a stretch for an obscure Internet subculture, unless I had a concrete example.

leogao 24 Feb 2026 8:43 UTC
11 points
1
in reply to: jefftk’s comment on: Storing Food
presumably people would not think to attack the random storage unit place though? like usually those are filled with random shit like furniture that’s useless in a famine. presumably very few people are putting food in their storage units. and the units are usually individually locked with pretty strong padlocks and made of pretty solid steel sheets, so a prospective robber would have to bring an angle grinder in and spend days slowly working their way through all of the units.
I guess the bigger concern is it might be hard to get safely from your apartment to the storage unit and back, if there’s violence in the streets?

leogao 24 Feb 2026 8:23 UTC
13 points
3
on: The persona selection model
for most of the examples under “empirical evidence for PSM”, you could have easily told an equally plausible story, if the evidence had come out the other way.
suppose emergent misalignment didn’t exist. well, clearly, because writing secure code is so hard, only a tiny fraction of insecure code in the wild is malicious. most people who write insecure code are simply uneducated in writing secure code, but are otherwise well meaning and upstanding members of civilization. so therefore when the model is trained on insecure code, PSM correctly predicts that the model will not become evil in other ways.
as a result, I think surprising forward predictions would be much more valuable than retrodictions.

leogao 24 Feb 2026 7:42 UTC
12 points
2
on: The ML ontology and the alignment ontology
fwiw, my impression at the time was weak to strong was explicitly intended to take alignment and frame it in the ML ontology, so that ML researchers could contribute to it. (in retrospect this didn’t pan out, imo somewhat predictably, though I wasn’t ultra confident it wouldn’t pan out.) also, in general, I think the final paper ended up predominately reflecting Collin’s vision, more so than Ilya’s or anyone else’s.
I’m still not convinced any of the persona stuff has produced anything of value. it appears to me to be unfalsifiable just so stories. as cynical as I am about typical lab safety work, I’ve come to believe that even typical lab safety is more positively impactful than the persona stuff (because in short timeline worlds, the alignedness of current models is more directly relevant to the base case of RSI). the persona people like to complain about how hamfisted the lab safety work is or whatever, but it’s always easier to complain from the sidelines.