--- Log opened Thu Dec 16 00:00:02 2021
00:59 < nsh> wake me up when they find a way to interfere fields atemporally
01:24 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has joined #hplusroadmap
01:58 -!- spaceangel [~spaceange@ip-78-102-216-202.net.upcbroadband.cz] has joined #hplusroadmap
02:30 -!- nmz787 [~nmz787@user/nmz787] has quit [Ping timeout: 256 seconds]
02:44 -!- dr-orlovsky [~dr-orlovs@31.14.40.18] has joined #hplusroadmap
02:48 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 240 seconds]
02:59 -!- nmz787 [~nmz787@user/nmz787] has joined #hplusroadmap
04:13 -!- CryptoDavid [uid14990@id-14990.uxbridge.irccloud.com] has joined #hplusroadmap
05:58 -!- dr-orlovsky [~dr-orlovs@31.14.40.18] has quit [Ping timeout: 240 seconds]
05:59 -!- dr-orlovsky [~dr-orlovs@31.14.40.18] has joined #hplusroadmap
06:13 -!- deltab [~deltab@user/deltab] has quit [Ping timeout: 250 seconds]
06:21 -!- deltab [~deltab@user/deltab] has joined #hplusroadmap
06:25 -!- nmz787 [~nmz787@user/nmz787] has quit [Ping timeout: 256 seconds]
08:54 < docl> .tw https://twitter.com/Ben_Reinhardt/status/1471523064416452610
08:54 < saxo> There's been an explosion of excitement around grants to enable new research, but grants can only go so far. // I wrote an essay outlining the limits of grants, why research management matters, and some potential alternatives. // https://benjaminreinhardt.com/grants (@Ben_Reinhardt)
09:11 < muurkha> how clueless do you have to be to "write an essay outlining the limits of ... grants to enable new research", not mention the NIH even once, mention the NSF only once in passing, and mention DARPA only implicitly and only in the context of the 01960s?
09:13 < muurkha> this is also pretty spectacularly clueless:
09:13 < muurkha> > Grants don’t require a huge time commitment or deep expertise on the part of the grantor (it’s pretty straightforward to get an experts take on any particular grant).
09:15 < muurkha> http://www.aaronsw.com/weblog/delegation describes the deep recursive problem with that approach: how do you find out who the experts are?  When George Washington was sick, how did he decide which doctors would treat him?
09:22 < muurkha> I suspect that Reinhardt is just completely unaware of the US$8 billion a year given out in grants by the US NSF and the US$26 billion a year given out in grants by the NIH
09:26 < fenn> "the 20-layer actuator, which requires less than 500 volts to operate" https://techxplore.com/news/2021-12-low-voltage-power-dense-artificial-muscles-microrobots.html
09:27 < muurkha> some of the things he lists in his "alternatives to grants" section are things that experienced granting agencies do routinely
09:30 < fenn> reinhardt is now running a "private DARPA" and is trying to justify why that's a good idea
09:31 < fenn> gah stupid case-sensitive urls
09:32 < fenn> caution: big page. https://benjaminreinhardt.com/parpa
09:33 < fenn> i forget what the organization is called but there was some crypto money involved
09:33 < fenn> http://astera.org
09:35 < fenn> my mom worked at NSF deciding which grants get funded, her job was primarily doing that expert-delegation thing
09:36 < fenn> the main problem was that they did not have enough money to give all the deserving grant requesters money, so inevitably you had to sort them into more and less deserving, which is subjective
09:38 < fenn> basically the procedure looks something like: look up related papers in the field, email the authors "who is an expert on X?" and then email the suggested expert with questions about the grant
09:40 < fenn> with a private grant agency you have freedom to narrow the set of grant requests to something that you have expertise in personally
09:48 < kanzure> https://there.oughta.be/a/wifi-game-boy-cartridge
09:48 < kanzure> why didn't he try booting a game over wifi
09:50 < fenn> yeah clearly it should have a catalog of ROMs that you download and then run
09:50 < fenn> i'm pretty impressed with the PCB rendering
09:55 < fenn> it seems obvious that he should sell them right? i'm not blind and missing the prominent "buy it now" link?
09:56 < kanzure> next up wifi64
09:57 < muurkha> fenn: you'd think that Reinhardt would talk to at least one NSF- or NIH-funded PI about the granting process in order to understand how it works
09:57 < muurkha> the problem with the procedure you're suggesting is that you end up with George Washington's doctor
09:58 < muurkha> or, in the case of nanotechnology, you end up with metallurgists and people studying medieval stained glass and Smalley rebranding their work as "nanotechnology" so they can slurp up those delicious Drexler grants
09:58 < fenn> i didn't say it's a good procedure, i was describing how it actually works at NSF
09:59 < muurkha> oh, sorry
09:59 < muurkha> agreed
09:59 < fenn> i haven't thought about it enough to know what's wrong with the procedure
10:00 < muurkha> I mean there are a lot of different things that go into NSF's procedure (and NIH's, and SBIR's, and DARPA's...)
10:00 < kanzure> how about: government is ungood at allocation of resources and nobody has invented a way of funding R&D that makes sense to everyone involved
10:00 < kanzure> besides patrons
10:00 < fenn> universities taking a huge cut of every grant has got to be a big one
10:01 < fenn> coupled with the implicit requirement that every grantee works at a university
10:01 < muurkha> yeah, the burgeoning managerial administrator cut since the 01970s has been a huge problem
10:01 < muurkha> not just for grant overhead but also for tuition
10:01 < muurkha> kanzure: with respect to research, I don't think government is particularly worse at it than anyone else
10:02 < fenn> i really have no idea where all the university money goes
10:02 < muurkha> I think there are some SSC articles about this?
10:02 < fenn> i would be happy with a pie chart, as long as it's accurate
10:02 < muurkha> I don't think it can be
10:03 < muurkha> markets, in particular, aren't very good at allocating resources to things that can't internalize their returns.  stock markets and futures markets in theory could improve this situation
10:03 < kanzure> where is paul sztorc when you need him
10:03 < fenn> or one of those flow network charts like https://en.wikipedia.org/wiki/File:Energy_Flow_US_2019.png
10:03 < kanzure> (disciple of robin hanson i guess)
10:04 < muurkha> one problem with that is that if you tell people why you are betting that a particular thing will happen, you lose your edge
10:04 < fenn> lol @truthcoin
10:04 < fenn> the conceit that markets can decide truth
10:04 < muurkha> so if one of your other bets pays off before it happens, you can't reinvest your winnings toward your other prediction
10:04 < fenn> (let's not get sidetracked again with truth philosophy plz)
10:04 < muurkha> haha
10:05 < fenn> please explain what the phrase "internalize their returns" means
10:05 < muurkha> sorry!
10:05 < muurkha> it means
10:06 < muurkha> that when the thing you did produces a benefit
10:06 < muurkha> that benefit flows to you
10:06 < fenn> what fraction of the benefit do you need to capture? anything > 0%?
10:06 < muurkha> rather than some random guy in Germany or Bangladesh
10:07 < muurkha> any fraction makes some ventures profitable, but higher fractions make more ventures profitable
10:07 < fenn> with a kickstarter, you can fund a project and then it is personally beneficial, but you don't necessarily capture the value and make it inaccessible to others
10:08 < fenn> is kickstarter a market?
10:08 < muurkha> well, you usually capture some of it.  people use kickstarter pretty often for preordering products
10:08 < muurkha> yes
10:09 < muurkha> and it's very plausible that dominant assurance contracts, which are very similar to kickstarter, can solve the public-goods problem for certain classes of public goods
10:09 < fenn> there are two types of kickstarters in common use; 1) "fund me to finish my comic book and then you can read it for free!" 2) buy a copy of the widget we have already designed and negotiated bulk widget production costs with the factory
10:09 < fenn> in 1) a rational agent would benefit from waiting for others to fund the project, and then read for free at no cost in total
10:10 < muurkha> things like the Fontus scam on Kickstarter demonstrate why it doesn't work well for applied product development
10:10 < fenn> what's that?
10:10 < muurkha> Fontus, the Self-Filling Water Bottle
10:10 < fenn> is it as dumb as it sounds?
10:10 < muurkha> well, a dehumidifier can be a useful source of water under some circumstances
10:11 < muurkha> but basically they promised to make a dehumidifier impossibly small and violating Carnot's limits, and by the time they realized that was what they'd done they were already on the hook to deliver it
10:12 < fenn> a lot of type 2 kickstarters have already demonstrated a working prototype
10:12 < muurkha> Fontus claimed they had one
10:12 < muurkha> I mean they filmed what they claimed was a working prototype
10:12 < muurkha> the problem is that Fontus had much better marketing videos than many actually useful projects
10:12 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has joined #hplusroadmap
10:13 < muurkha> and the marketing videos are what backers see, not the underlying science
10:14 < muurkha> Kickstarter is lacking even the 'look up related papers in the field, email the authors "who is an expert on X?" and then email the suggested expert with questions about the grant' function
10:14 < fenn> yes
10:14 < fenn> kickstarter needs a "critcism" section
10:15 < muurkha> I'm not sure that would help
10:15 < muurkha> then they would just fund the projects whose competitors had the worst lawyers
10:15 < fenn> lawyers is not really what i'm imagining providing the criticism
10:16 < muurkha> lawyers are the people who spend their lives providing the most persuasive possible criticisms
10:16 < fenn> i don't know how to solve this problem without turning the world into an inescapable ministry of truth and trust dystopia
10:16 < muurkha> ultimately you need expertise in the field to judge which grants are worth funding, and you need either expertise in the field or empirical results that are obvious even to the vulgar and ignorant
10:16 < muurkha> and sometimes even that doesn't help because then the vulgar and ignorant have to guess what to attribute the results to
10:17 < muurkha> uh, missing part of second clause: you need either expertise or results to recognize expertise in the field
10:17 < fenn> yes, sorry, i'm not going to propose solutions to the delegation of expertise problem
10:18 < fenn> i want to maintain the fiction that investors are competent at discerning value
10:18 < muurkha> DARPA does "program reviews" and "site visits", not sure what the NSF equivalent is
10:18 < fenn> much less hands on
10:18 < muurkha> investors are generally competent at discerning *captured* value, not *produced* value
10:19 < muurkha> internalities rather than externalities
10:19 < fenn> hmm. well as long as everything boils down to dollar amounts that's true
10:20 < muurkha> for things that don't boil down to dollar amounts, like prestige, pleasure, and wisdom, it's sometimes even easier, though scammers frequently enough pass off the illusion of wisdom as wisdom
10:21 < muurkha> by saying mysterious things
10:22 < muurkha> if I invest US$10000 in fuzzing a PDF parser, I don't know what the results will be until the research is done, and it may not find anything
10:22 < muurkha> (fwiw I'm nominally working on a PDF-parser-related program for DARPA atm)
10:23 < fenn> but at the end you will have advanced the state of PDF fuzzing by several orders of magnitude
10:23 < fenn> probably more useful than calculating the digits of pi!
10:23 < muurkha> no, not even close
10:23 < muurkha> way more money than that has already gone into PDF fuzzing
10:23 < fenn> what really, like, hardware costs? or labor costs?
10:23 < muurkha> even on just the program I'm currently working on, in fact
10:23 < muurkha> both
10:24 < muurkha> if I invest those US$10000 in a marble countertop instead, to give me pleasure when I'm in my kitchen, I can have a pretty good idea of what it will look like
10:24 < fenn> is this a task where you just plug in PCs in parallel and let them run on a source of cheap electricity?
10:24 < muurkha> and six weeks later when the kitchen remodeling is done I can start to enjoy the counter
10:24 < muurkha> in a way that doesn't boil down to dollar amounts
10:25 < fenn> it's a lottery
10:25 < muurkha> what, fuzzing or kitchen remodeling?
10:25 < fenn> in fuzzing you buy a ticket for 1e-12 USD and maybe you win a new bug
10:25 < muurkha> ah, sure
10:25 < muurkha> but it's not primarily that
10:26 < muurkha> lots of PCs in parallel on a source of cheap electricity is just a part of it
10:26 < muurkha> if you feed chunks from /dev/urandom to PDF parsers you will not find anything
10:26 < muurkha> you have to be more intelligent about the searching process
10:26 < fenn> i'm sure there are lots of fancy heuristics and sorting through the discovered bugs for interestingness
10:27 < muurkha> and then you have to minimize the results, and sort them into equivalence classes
10:27 < fenn> what's more valuable in expected bug output, $1 of programmer time, or $1 of fuzzing
10:27 < muurkha> because you may find 1e6 PDFs that trigger crashes
10:27 < fenn> right right
10:27 < muurkha> and three of them are for one bug and all the others are for a second bug
10:27 < muurkha> $1 of programmer time, because every increment in heuristics produces an exponential speedup in the search process
10:28 < fenn> so we're basically not limited by fuzzing CPU time at all, yes?
10:28 < muurkha> and not only at that moment but thereafter
10:28 < muurkha> no, because without fuzzing CPU time you can't do any fuzzing
10:28 < muurkha> adding more computation provides a linear speedup
10:28 < fenn> it's not a limiting factor
10:29 < fenn> i'm not saying you can do fuzzing with 0 cpu cycles
10:29 < muurkha> it is, because you can't arbitrarily add programmers to the project and improve your heuristics arbitrarily fast
10:29 < fenn> hmm
10:30 < muurkha> you're bottlenecked on the people you can recruit, your ability to assess their ability, their ability to learn, coordination problems between programmers, etc.
10:30 < fenn> you'll have a better idea of what the marble countertop will be like than the recruited programmers
10:31 < muurkha> exactly.  and a fortiori for the bugs found by the recruited programmers
10:31 < fenn> i thought you were complaining about the randomness of the fuzzing process
10:31 < muurkha> no, the randomness of the research process.  and not exactly complaining
10:32 < muurkha> just pointing at why it's more difficult for investors to discern the future value produced by funding research
10:33 < muurkha> in our case, because we're working for DARPA and aiming to fix bugs rather than exploit them, the financial returns for us are pretty predictable
10:33 < muurkha> and for DARPA
10:34 < muurkha> there seem to be actors out there for which this is not the case because their potential winnings are, say, the profits from Azure and AWS, both of which had outages yesterday
10:43 < muurkha> as Hobbes pointed out, this is not a stable situation; once four different ransomware raiders are each extorting 30% of AWS's profits, AWS runs at a loss
13:07 < kanzure> speaking of PARPA.. and FROs.. https://mobile.twitter.com/heyjudka/status/1471245655595118594
13:08 < muurkha> .t
13:08 < saxo> 2021: year of the Modern Science Revolution? 👩‍🔬🧪 // A cambrian explosion of new science funding models was driven by covid urgency, open science, frustration w/status quo. Most exciting: these models are built for translation & startup creation! // 🧵of new science funding models: (@heyjudka)
13:09 < muurkha> we'll see!
13:10 < muurkha> my concern with startup creation is that the knowledge ratchet of scientific progress only applies to published knowledge, and our current incentives for startups seem to strongly punish publishing their knowledge
13:12 -!- nmz787 [~nmz787@user/nmz787] has joined #hplusroadmap
14:51 -!- CryptoDavid [uid14990@id-14990.uxbridge.irccloud.com] has quit [Quit: Connection closed for inactivity]
15:17 -!- spaceangel [~spaceange@ip-78-102-216-202.net.upcbroadband.cz] has quit [Remote host closed the connection]
16:30 < fenn> "In order to use certain classifier constructions, the signer must be able to visualize the entity and its shape, orientation and location. It has been shown that deaf signers are better at generating spatial mental images than hearing non-signers. The spatial memory span of deaf signers is also superior. This is linked to their use of sign language, rather than being deaf"
16:34 -!- nmz787 [~nmz787@user/nmz787] has quit [Ping timeout: 240 seconds]
16:49 -!- nmz787 [~nmz787@user/nmz787] has joined #hplusroadmap
18:44 -!- Malvolio [~Malvolio@user/malvolio] has quit [Killed (molybdenum.libera.chat (Nickname regained by services))]
18:45 -!- Malvolio [~Malvolio@user/malvolio] has joined #hplusroadmap
21:34 < lsneff> The semester just finished for me
21:34 < lsneff> It’s been a tough couple months
22:02 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 245 seconds]
22:29 -!- nmz787 [~nmz787@user/nmz787] has quit [Ping timeout: 240 seconds]
23:03 < maaku> fenn: related is that austronesian tribe that only uses absolute directions (e.g. north/south/east/west) and not relative directions (left/right)
23:03 < maaku> surprise surprise, those tribemembers have a really good internal mental compass
23:04 < maaku> (but not the tribe members who grew up monolingual English)
23:04 < maaku> (weak) Sapir–Whorf hypothesis
23:05 < maaku> one of the reasons I really want a learnable logical language
23:08 < fenn> i think in both cases they just use the relevant skill much more often and so get better at it
23:20 < maaku> fenn: in the individual case, yes
23:21 < maaku> but I'd like to live in a society where people are on the whole less stupid and less easily manipulatable
23:23 < maaku> I hypothesize that you could raise the bar of rationality by having people's native/primary tongue be a monoparsable logical language
23:24 < maaku> of course supplanting English is a boil-the-ocean scale project ... but if the language was simple and easy enough to learn, it is possible it could get traction as on auxlang in an entirely new frontier like Mars
23:25 < maaku> the existing options (Lojban and Toaq) really suck though
23:26 < maaku> actually Toaq would be okay, but not if you're assuming people came from any of the major spoken languages except maybe chinese
--- Log closed Fri Dec 17 00:00:03 2021