at the end of this article the interviewer clarifies that she verified this to be SBF, but does not clarify that SBF understood this conversation to be public (which suggests he may not have). hoping for some clarification there, because it’s relevant to understanding the broader ethical context in which this is all playing out under.
ponkaloupe
Karma: 131
the second-order effects of turning off the WiFi are surely comprised of both positive and negative effects, and i have no idea which valence it nets out to.
these days homes contain devices whose interconnectedness is used to more efficiently regulate power use. for example, the classic utility-controlled water heater, which reduces power draw when electricity is more expensive for the utility company (i.e. when peakers would need to come online). water heaters mostly don’t use WiFi but thermostats like Nest, programmable light bulbs, etc do: when you disrupt that connection, in which direction is power use more likely to change?
i have my phone programmed so that when i go to bed (put it in the “i’m sleeping” Do Not Disturb mode) it will automatically turn off all the outlets and devices — like my TV, game consoles, garage space heater — which i only use during the day. leaving any one of these on for just one night would cancel weeks of gains from disabling WiFi.
i’m naive to the details of GPT specifically, but it’s easy to accidentally make any reduction non-deterministic when working with floating point numbers — even before hardware variations.
for example, you want to compute the sum over a 1-billion entry vector where each entry is the number 1. in 32-bit IEEE-754, you should get different results by accumulating linearly (1+(1+(1+…))) vs tree-wise (…((1+1) + (1+1))…).
in practice most implementations do some combination of these. i’ve seen someone do this by batching groups of 100,000 numbers to sum linearly, with each batch dispatched to a different compute unit and the 10,000 results then being summed in a first-come/first-serve manner (e.g. a queue, or even a shared accumulator). then you get slightly different results based on how each run is scheduled (well, the all-1’s case is repeatable with this method but it wouldn’t be with real data).
and then yes, bring in different hardware, and the scope broadens. the optimal batching size (which might be exposed as a default somewhere) changes such that even had you avoided that scheduling-dependent pitfall, you would now see different results than on the earlier hardware. however, you can sometimes tell these possibilities apart! if it’s non-deterministic scheduling, the number of different outputs for the same input is likely higher order than if it’s variation strictly due to differing hardware models. if you can generate 10,000 different outputs from the same input, that’s surely greater than the number of HW models, so it would be better explained by non-deterministic scheduling.
OpenAI estimated that the energy consumption for training GPT-3 was about 3.14 x 10^17 Joules.
sanity checking this figure: 1 kWh is 1000 x 60 x 60 = 3.6 MJ. then GPT-3 consumed 8.7 x 10^10 kWh. at a very conservative $0.04/kWh, that’s $3.5B just in the power bill — disregarding all the non-power costs (i.e. the overheads of operating a datacenter).
i could believe this number’s within 3 orders of magnitude of truth, which is probably good enough for the point of this article, but i am a little surprised if you just took it 100% at face value.
Not only do our brains say “no fuck you, you don’t get to work on rockets”
getting yourself to somewhere on this curve which is not the far left but also not too far to the right can be unbelievably productive. there’s a certain type of infatuation which drives one to show off their achievements, which in turn requires one to make achievements. building a rocket, and inventiveness in general, is a decently high status thing: you may experience a greater drive to actually do these things during a certain period of infatuation.
an important aspect of Pomodoro at a corporate gig is not “how long can i remain attentive” but “how long is it acceptable to be unreachable for”. it’s about guaranteeing yourself an uninterrupted chunk of time, by disabling slack/email/etc. as long as you’re doing vaguely productive things for that time slot, you’ve already unlocked most of the benefits.
for truly personal work, i mostly don’t use Pomodoro timers, and go for a more freeform approach: once i feel myself slowing down, i’ll set down my work, take a break, and then pick a different thing off my to-do list. the exception is for tasks that i have trouble getting started on. say, a book whose opening isn’t hooking me. i’ll set a timer promising to do the activity for 15 minutes. during those minutes, i free myself from thinking about the meta picture of “is this the right task to be working on” and can properly focus on getting into the book. when the alarm goes off, only then do i reconsider my priorities, and either keep at the task without the timer, or put it down.
in the end, maybe it’s just about being aware/explicit with your time. no timer is going to force you into flow. but it will force you to think more critically about your time.
interoperability. we take it forgranted everywhere else in life: when you have to replace a fridge it’s easy because they all have the same electrical/water hookups. replace a door, same thing: standardized size, hinges, and knobs. going further, i’ve been upgrading the cabinets/drawers in my kitchen: they’re standard size so i can buy 3rd party silverware inserts, or even inserts made specifically to organize anything that’s k-cup shaped. i replaced the casters on my office chair with oversized carpet-friendly wheels: standardized attachments. so many things in the physical world are made to be interoperable because it facilitates mass production and allows for any company to innovate in any sliver they see. it’s cheaper for producers, and improves the consumer experience.
i assure you those causes and benefits aren’t restricted to the physical world. i read this post in my RSS client, even as my roommate was fiddling with the router because all my RSS feeds get saved for offline reading in the background, before i even decide to read them. simultaneously that RSS standard allows LessWrong to get more reach.
i confront the crux of your post differently: “how do i navigate adversarial relationships (with a business)”? increasingly my approach is to just not engage (or engage less). when it comes to mid-size group stuff, it’s usually pretty easy: LW is just better than Facebook, reddit, or anything that sees its users as a resource to extract from.
for smaller groups or1-to-1 things i choose SMS over Discord; for the people where that’s too low-bandwidth and IRL hangouts aren’t practical, treat any monopoly replacement (signal, telegram, etc) as explicitly ephemeral: as these services switch to value capture we hop ship without losing anything. the world is large enough that there are plenty of substitute activities even if you disengage from Facebook, say. but it’s easier to adopt a policy of “don’t engage” a priori, rather than integrate them into your life and then decide to cut back on them..
On the other hand, every known living creature on Earth uses essentially the same DNA-based genetic code, which suggests abiogenesis occurred only once in the planet’s history.
well this alone doesn’t suggest abiogenesis occurred only once: just that if any other abiogenesis occurred it was outcompeted by DNA replicators.
when i was in school there was a theory that RNA was a remnant of pre-DNA abiogenesis: either that it bootstrapped DNA life, or that it was one such distinct line which “lost” to DNA. in the latter case, hard to say how many other lines there were which left no evidence visible today, or even how many lines/abiogenesis events would have occurred if not for DNA replicators altering the environment and available resources. hopefully research has provided better predictions here that i just haven’t heard about yet.
regarding that poll:
[how much would you pay] to go back to the way of life you had before the pandemic?
i really can’t think of any previous lifestyle that i actually want to go back to living. maybe it’s because i’m young, but if my lifestyle this year is ever worse than the prior year, that’s pretty close to saying that i failed at life for that year. humans are adaptable: i’m not at all surprised if 42% of respondents have adapted over the last two years and created new habits and ways of life that they aren’t interested in giving up. i’m \it{happy} for those who wouldn’t revert their lifestyle changes for a modest amount of $, because it hints that they made very positive lifestyle changes.
this does help the original question: “where is everybody” can reasonably be answered with “they’re on the other side of a coin flip”. in the point estimate version it was “they’re on the other side of some hundreds of consecutive coin flips”. so it helps the original question in that there’s far less that needs to be explained.
If people were more aware of the limits of politics, disengagement and cynicism would probably increase. These attitudes are already a problem, particularly among the less educated, and are associated with a series of negative outcomes.
this is a particularly interesting statement to me. on the one hand, the bulk of your post is about illuminating the limits of politics, and you mention academics and such admitting to these limits. hence, “awareness of the limits of politics” is supposed to be a highly-educated view. then you illustrate the downstream effects of disengagement and cynicism, but cite these as problems among the lesser-educated — the opposite end of the education spectrum.
so which is it? is awareness of the limits of politics a good thing only when a person is “highly educated enough” to convince themselves that carrying on the illusion, and not disengaging, is a good thing? if this is true, you should be able to convince the other person of this view, and then feel safe in revealing the rest of the truth. much of this post has the vibe of “we can’t trust these fools with the truth, thus we’re going to withhold it and thereby ensure that they remain fools”… there’s a lot of hubris to that idea.
You explain how it wouldn’t be costly (to people who can afford backpacks), then insist it would be costly.
when i was a kid, i’d bring an umbrella with me to the bus stop on rainy days. when i boarded the bus, i’d set it down next to me to give it time to dry off (didn’t want to get the books inside my pack wet). i can’t tell you how many trips i had to make to the lost and found after forgetting that umbrella on the bus, in the lunchroom, or anywhere else.
in the end, i ditched the umbrella and waited for the bus under the overhang of my neighbor’s porch. umbrellas just aren’t worth the hassle for all but the worst of storms in the PNW — backpack or no. well, unless the rain really bothers you or you’re exceptionally unforgetful.
in an earlier, social, life, i always met my dates through my friends. i never had to go out of my way to make friends of both sexes, my hobbies/interests just happened to attract both. 10 years later, all my friends are dudes, 90% of my friends’ friends are dudes, and my social ties are just fewer in number.
i’m focused on just growing my immediate friend network at the moment. i was raised to not use my real name on the internet, much less post photos of my face on it. i’ve had just enough continued success in that approach that i haven’t been forced to concede. but i legitimately enjoyed having my friends set me up on dates. i’m not ready to abandon that method of dating and relegate matchmaking to some app behind a screen. part of this is the expectation that i’m far less likely to be compatible with the wide pool of candidates on a dating app than with the narrower pool filtered through my lifelong friends. but a bigger part is probably just stubbornness, holding onto a dying model. and maybe some amount of risk aversion — not of the “fear of rejection” type — but of dismissing the new models without first risking them. i think it’s distinct from laziness, but maybe it falls into #9.
Instincts to punish people are how actual humans precommit.
i think you could equally frame this as “people precommit due to an expectation of reciprocity”. like, i don’t generally follow through on my commitments to plans with friends because i fear punishment for breaking them. it’s more that i expect whatever amount i invest into the friendship will be reciprocated (approximately).
you could frame the fallout of a commitment failure as “punishment”, but if the risk of punishment exceeded the benefit of cooperation that would discourage me from pre-committing; from interacting with the thing at all. if i thought my crush would beat me should i break things off with him, then i’d simply never ask him out to begin with and we’d probably both be worse off for that.
The underground bunker is only weird if I’m vocal about it. Am I posting this from behind tor? Do I pass my messages through an anti-stylometry filter before sharing them? I’m already operating behind a nonsense pseudonym, but if I used a real-sounding name that wasn’t my legal name, would you question it? It’s often the case that effective privacy techniques simply aren’t easily detectable by 3rd or even 2nd parties: the goal is frequently to blend in (to become a part of some larger “anonymity set”).
you cite software. turnover in that industry tends to be pretty high, which means the successful companies end up with some structure where when any particular engineer leaves, their responsibilities are re-distributed quickly. so (1) if a team member is out for a week their most urgent responsibilities can already be picked up by their peers and (2) wherever this isn’t the case, the employer is likely to see that as a failure in need of fixing.
unaligned week-long holidays sort of serve as a dry run to ensure that things are set up such that when the employee on leave were to leave permanently things would still continue.
there’s another way to address this coordination, btw: ask that employees schedule their time away 2 weeks in advance: when the team plans the next chunk of work, sequence things such that this employee won’t be working on a blocking task near the interval where they’re away.
regardless, certain Schelling points do arise naturally. it’s far more common for someone to take individual Fridays off than it is to take Tuesdays off, for example. weekends themselves are some example of what you’re getting at: extending those (by occasionally adding Fridays or Mondays) seems an easy course to follow.
Why do you think my post is being shot down?
my first instinct is general “politics is the mindkiller” weariness, but i wonder if it’s more an issue of scope. it’s framed to be relevant only to Floridians, and relevant answers would have to be very broad with little depth (“vote A, B, … and Z”) or deep but tangential and not a direct answer (“i like candidate Q because of their proposed policy R which is good because S but has some uncertainties around T…”).
it also feels like you’re offloading too much to the reader. it’s easily mistaken as a “do my homework for me” request, even. i have no idea what’s on the ballot, and i guess if i were in Florida i’m supposed to fish out the ballot and study up on it first? just hope i happen to be near my desk or i’ll have to google around for an online version.
if you want, you might get more discussion by taking one policy on the ballot, decoupling it from the specific geography, and then identifying a few intriguing ramifications/uncertainties as starting points for a (more focused) discussion. food for thought, as i’m in no position to speak for all LWers.
this could make certain statistical measurements less noisy, but as you point out there are so many confounding variables to deal with (e.g. period effects). if we couldn’t conclude anything from the 50 years ago where we made this same change (in reverse), i don’t quite understand what’s different this time around such that we will be able to conclude things from this change.
there’s a lot of ways you can lose a domain. ICANN requires a domain’s WHOIS records (which includes email, tel, and physical address) to be accurate. i don’t have much experience with enforcement, but i think some TLDs are more impacted by this than others — e.g. .us explicitly treats WHOIS records as public and periodically “spot checks” the accuracy of records. [1]
additionally, ownership over the popular .io TLD has been contested in the past. i understand that the DNS root servers themselves are highly decentralized, across continents even, but the smaller TLDs and the registration part of it feels like it might be one of those human systems that relies heavily on norms and pure-hearted authorities.
1: https://www.washingtonpost.com/wp-dyn/articles/A7251-2005Mar4.html
2: https://salt.agency/blog/biot-chagos-islands-risk-to-io-cctld-domains/
I’m relatively OOTL on AI since GPT-3. My friend is terrified that we need to halt it urgently: I couldn’t understand his point of view; he mentioned this book to me. I see a number of pre-readers saying the version they read is well-suited exactly for convincing people like me. At which point: if you believe the threat is imminent, why delay the book four months? I’ll read a digital copy today if you point me to it.