Posts

Garrabrant and Shah on human modeling in AGI 2021-08-04T04:35:11.225Z
Finite Factored Sets: LW transcript with running commentary 2021-06-27T16:02:06.063Z
"Existential risk from AI" survey results 2021-06-01T20:02:05.688Z
Predict responses to the "existential risk from AI" survey 2021-05-28T01:32:18.059Z
Sabien on "work-life" balance 2021-05-20T18:33:37.981Z
MIRI location optimization (and related topics) discussion 2021-05-08T23:12:02.476Z
Scott Alexander 2021 predictions: calibration and updating exercise 2021-04-29T19:15:39.463Z
Thiel on secrets and indefiniteness 2021-04-20T21:59:35.792Z
2012 Robin Hanson comment on “Intelligence Explosion: Evidence and Import” 2021-04-02T16:26:51.725Z
Logan Strohl on exercise norms 2021-03-30T04:28:22.331Z
Julia Galef and Matt Yglesias on bioethics and "ethics expertise" 2021-03-30T03:06:07.323Z
Thirty-three randomly selected bioethics papers 2021-03-22T21:38:08.281Z
Politics is way too meta 2021-03-17T07:04:42.187Z
Deflationism isn't the solution to philosophy's woes 2021-03-10T00:20:07.357Z
What I'd change about different philosophy fields 2021-03-08T18:25:30.165Z
MIRI comments on Cotra's "Case for Aligning Narrowly Superhuman Models" 2021-03-05T23:43:54.186Z
Utilitarian doppelgangers vs. making everything smell like bananas 2021-02-20T23:57:34.724Z
MIRI: 2020 Updates and Strategy 2020-12-23T21:27:39.206Z
Cartesian Frames Definitions 2020-11-08T12:44:34.509Z
"Cartesian Frames" Talk #2 this Sunday at 2pm (PT) 2020-10-28T13:59:20.991Z
Updates and additions to "Embedded Agency" 2020-08-29T04:22:25.556Z
DontDoxScottAlexander.com - A Petition 2020-06-25T05:44:50.050Z
EA Forum AMA - MIRI's Buck Shlegeris 2019-11-15T23:27:07.238Z
A simple sketch of how realism became unpopular 2019-10-11T22:25:36.357Z
Christiano decision theory excerpt 2019-09-29T02:55:35.542Z
Kohli episode discussion in 80K's Christiano interview 2019-09-29T01:40:33.852Z
Rob B's Shortform Feed 2019-05-10T23:10:14.483Z
Helen Toner on China, CSET, and AI 2019-04-21T04:10:21.457Z
New edition of "Rationality: From AI to Zombies" 2018-12-15T21:33:56.713Z
On MIRI's new research directions 2018-11-22T23:42:06.521Z
Comment on decision theory 2018-09-09T20:13:09.543Z
Ben Hoffman's donor recommendations 2018-06-21T16:02:45.679Z
Critch on career advice for junior AI-x-risk-concerned researchers 2018-05-12T02:13:28.743Z
Two clarifications about "Strategic Background" 2018-04-12T02:11:46.034Z
Karnofsky on forecasting and what science does 2018-03-28T01:55:26.495Z
Quick Nate/Eliezer comments on discontinuity 2018-03-01T22:03:27.094Z
Yudkowsky on AGI ethics 2017-10-19T23:13:59.829Z
MIRI: Decisions are for making bad outcomes inconsistent 2017-04-09T03:42:58.133Z
CHCAI/MIRI research internship in AI safety 2017-02-13T18:34:34.520Z
MIRI AMA plus updates 2016-10-11T23:52:44.410Z
A few misconceptions surrounding Roko's basilisk 2015-10-05T21:23:08.994Z
The Library of Scott Alexandria 2015-09-14T01:38:27.167Z
[Link] Nate Soares is answering questions about MIRI at the EA Forum 2015-06-11T00:27:00.253Z
Rationality: From AI to Zombies 2015-03-13T15:11:20.920Z
Ends: An Introduction 2015-03-11T19:00:44.904Z
Minds: An Introduction 2015-03-11T19:00:32.440Z
Biases: An Introduction 2015-03-11T19:00:31.605Z
Rationality: An Introduction 2015-03-11T19:00:31.162Z
Beginnings: An Introduction 2015-03-11T19:00:25.616Z
The World: An Introduction 2015-03-11T19:00:12.370Z

Comments

Comment by Rob Bensinger (RobbBB) on Garrabrant and Shah on human modeling in AGI · 2021-08-04T16:02:47.418Z · LW · GW

Edited!

Comment by Rob Bensinger (RobbBB) on Jimrandomh's Shortform · 2021-07-26T22:20:42.418Z · LW · GW

Prev/next is probably good enough.

Comment by Rob Bensinger (RobbBB) on Jimrandomh's Shortform · 2021-07-25T19:11:33.440Z · LW · GW

These nutrition posts are great.  Will there be a way for me to link to all (and only) this series, in chronological order, at some point? I want these discussed as a group on social media and the EA Forum too.

Comment by Rob Bensinger (RobbBB) on Jimrandomh's Shortform · 2021-07-25T19:03:27.760Z · LW · GW

My understanding is that diet RCTs generally show short-term gains but no long-term gains. Why would that be true, if the Regression to the Mean Diet is the main thing causing these results? I'd have expected something more like 'all diets work long-term' rather than 'no diets work long-term' from the model here.

Comment by Rob Bensinger (RobbBB) on Book review: The Explanation of Ideology · 2021-07-21T23:34:33.291Z · LW · GW

What's the definition of universalism? It's mentioned a bunch in the post.

Comment by Rob Bensinger (RobbBB) on "If and Only If" Should Be Spelled "Ifeff" · 2021-07-16T23:10:47.160Z · LW · GW

What about "if(f)"? Pronounced 'ifeff', but spelled 'if(f)' so it's both:

  1. Easy to tell that this is a variant of 'if', and if you round things off to colloquial 'if' then you'll at least sort of get what's being said (context probably helps).
  2. Easy to tell that it's not just 'if' or a typo for 'if'.
Comment by Rob Bensinger (RobbBB) on "If and Only If" Should Be Spelled "Ifeff" · 2021-07-16T23:08:15.124Z · LW · GW

Ahh, OK. This has the advantage that it's connected to 'iff', so you're competing less with existing usage.

Comment by Rob Bensinger (RobbBB) on "If and Only If" Should Be Spelled "Ifeff" · 2021-07-16T22:13:33.630Z · LW · GW

I think the reason "iff" hasn't caught on more broadly is because it's easily misread as "if"

Also, if you aren't familiar with "iff" you might see it and assume it's just a typo for "if". Jargon catches on better if it's obviously jargon, so people can google it or ask questions when they notice they don't understand.

Why 'ifeff', though? I'm not seeing a logical connection between the '-eff' and '... and only if'. I'd have expected the best shorthand to look more connected to the original, like 'iffo'.

Comment by Rob Bensinger (RobbBB) on This Sunday, 12PM PT: Scott Garrabrant on "Finite Factored Sets" · 2021-07-01T01:16:56.283Z · LW · GW

Here's the recording and transcript: https://www.alignmentforum.org/posts/6t9F5cS3JjtSspbAZ/finite-factored-sets-lw-transcript-with-running-commentary 

Comment by Rob Bensinger (RobbBB) on MIRIx Part I: Insufficient Values · 2021-07-01T00:39:03.837Z · LW · GW

I consider the Arbital article on CEV the best reference for the topic. It says:

CEV is rather complicated and meta and hence not intended as something you'd do with the first AI you ever tried to build. CEV might be something that everyone inside a project agreed was an acceptable mutual target for their second AI. (The first AI should probably be a Task AGI.)

So MIRI doesn't focus on CEV, etc. because the world hasn't nailed down step one yet. We're extremely worried that humanity's on track to fail step one; and it doesn't matter how well we do on step two if we don't pull off step one. That doesn't mean that stopping at step one and never shooting for anything more ambitious would be acceptable; by default I'd consider that an existential catastrophe in its own right.

Comment by Rob Bensinger (RobbBB) on Taboo "Outside View" · 2021-06-21T11:12:19.108Z · LW · GW

I like 'ass number' because it points at the actual experience / cognitive process behind these numbers. 'Wild guess' is vaguer -- e.g., if I'm using a standard statistical technique to estimate a number from other (observed) numbers, then I wouldn't call that an 'ass number', but I might still call it a 'wild guess' if the output is extremely uncertain.

Comment by Rob Bensinger (RobbBB) on Rob B's Shortform Feed · 2021-06-08T00:49:35.133Z · LW · GW

Shared with permission, a google doc exchange confirming Eliezer still finds the arguments for alignment optimism, slower takeoffs, etc. unconvincing:

Daniel Filan: I feel like a bunch of people have shifted a bunch in the type of AI x-risk that worries them (representative phrase is "from Yudkowsky/Bostrom to What Failure Looks Like part 2 part 1") and I still don't totally get why.

Eliezer Yudkowsky: My bitter take:  I tried cutting back on talking to do research; and so people talked a bunch about a different scenario that was nicer to think about, and ended up with their thoughts staying there, because that's what happens if nobody else is arguing them out of it.

That is: this social-space's thought processes are not robust enough against mildly adversarial noise, that trying a bunch of different arguments for something relatively nicer to believe, won't Goodhart up a plausible-to-the-social-space argument for the thing that's nicer to believe.  If you talk people out of one error, somebody else searches around in the space of plausible arguments and finds a new error.  I wasn't fighting a mistaken argument for why AI niceness isn't too intractable and takeoffs won't be too fast; I was fighting an endless generator of those arguments.  If I could have taught people to find the counterarguments themselves, that would have been progress.  I did try that.  It didn't work because the counterargument-generator is one level of abstraction higher, and has to be operated and circumstantially adapted too precisely for the social-space to be argued into it using words.

You can sometimes argue people into beliefs.  It is much harder to argue them into skills.  The negation of Robin Hanson's rosier AI scenario was a belief.  Negating an endless stream of rosy scenarios is a skill.

Caveat: this was a private reply I saw and wanted to share (so people know EY's basic epistemic state, and therefore probably the state of other MIRI leadership). This wasn't an attempt to write an adequate public response to any of the public arguments put forward for alignment optimism or non-fast takeoff, etc., and isn't meant to be a replacement for public, detailed, object-level discussion. (Though I don't know when/if MIRI folks plan to produce a proper response, and if I expected such a response soonish I'd probably have just waited and posted that instead.)

Comment by Rob Bensinger (RobbBB) on "Existential risk from AI" survey results · 2021-06-04T00:58:32.697Z · LW · GW

One-off, though Carlier, Clarke, and Schuett have a similar survey coming out in the next week.

Comment by Rob Bensinger (RobbBB) on "Existential risk from AI" survey results · 2021-06-02T19:10:37.842Z · LW · GW

Thanks, Josh!

Comment by Rob Bensinger (RobbBB) on Predict responses to the "existential risk from AI" survey · 2021-05-29T20:25:37.877Z · LW · GW

I've added six prediction interfaces: two for your own answers to the two Qs, two for your guess at the mean survey respondent answers, and two for your guess at the median respondent answers.

Comment by Rob Bensinger (RobbBB) on Predict responses to the "existential risk from AI" survey · 2021-05-29T16:33:29.140Z · LW · GW

I think it might be more interesting to sketch what you expect the distribution of views to look like, as opposed to just giving a summary statistic. I can add probability Qs, but I avoided it initially so as not to funnel people into doing the less informative version of this exercise.

Comment by Rob Bensinger (RobbBB) on This Sunday, 12PM PT: Scott Garrabrant on "Finite Factored Sets" · 2021-05-23T21:16:09.975Z · LW · GW

Here's the basic content, in the form of a transcript+video of a version of this talk Scott gave at Topos 12 days ago: https://www.lesswrong.com/posts/N5Jm6Nj4HkNKySA5Z/finite-factored-sets. This heavily overlaps with the LW talk today.

I'm guessing we'll release the LW talk video sometime too.

Comment by Rob Bensinger (RobbBB) on Peekskill Lyme Incidence · 2021-05-22T16:15:35.846Z · LW · GW

https://www.motherjones.com/politics/2017/09/dogs-can-be-vaccinated-against-lyme-disease-humans-cant/ says:

The immunization you can now give your puppy is essentially this original [LYMErix] vaccine, says Stanley Plotkin, a professor and consultant who literally wrote the book on vaccines, and whose son almost died from cardiac Lyme disease.

But we'd need a lot more detail and confirmation than just 'a second-hand claim from an expert that they're "essentially" the same'.

Comment by Rob Bensinger (RobbBB) on Sabien on "work-life" balance · 2021-05-21T23:10:21.648Z · LW · GW

Duncan posted it in a private chat I was in (and on FB); I asked if he'd cross-post to LW, or if he'd prefer that I post it for him; he voted for the latter. He didn't ask for it to be cross-posted, but also didn't object.

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-21T00:10:05.570Z · LW · GW

Thanks for the suggestion! Not sure why you can't contact me on LinkedIn, but I check email (and LW PM) more anyway; here's my email.

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-21T00:06:22.028Z · LW · GW

Thanks for the suggestions! SLC didn't make our top 30 list, but it's maybe top 50, and does seem like it has a lot of nice aspects!

From another comment I left here:

Research Triangle Park is on the list of 14+ cities that Alex classified one tier below our top thirty (as of one month ago): "Moved to backburner; had varying levels of initial interest but also flags; investigated and deprioritized; plausible something could bump them back onto our radar, but seems unlikely." Along with places like Ithaca NY, Salt Lake City UT, Pittsburgh / CMU, Princeton NJ, and Providence RI.

Comment by Rob Bensinger (RobbBB) on This Sunday, 12PM PT: Scott Garrabrant on "Finite Factored Sets" · 2021-05-19T13:21:30.382Z · LW · GW

Yes, and we'll be posting the basic content to LW later that day, so people who don't like talks or have scheduling conflicts can just read a written version.

Comment by Rob Bensinger (RobbBB) on New Magic Post: The Pro Tour Was Magic · 2021-05-19T01:22:26.576Z · LW · GW

A pro league with a rule like 'the maximum age is 28' might get the best of both worlds.

Comment by Rob Bensinger (RobbBB) on Alex Flint on "A software engineer's perspective on logical induction" · 2021-05-19T00:51:29.872Z · LW · GW

I was also hoping to see a recording -- if you give the talk again, or have a conversation like this, I'd be interested to check it or a recording out.

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-12T00:48:13.480Z · LW · GW

(Caveat: I’m mostly just replying to 'how many people do you need to fit?' here and discussing size, rather than trying to list everything we might want from indoor spaces. Also, my whole comment is simplifying things a lot, because a full answer would be too in-the-weeds. I err on the side of being way, way too concrete and hypothesis-privileging in order to paint an easy-to-visualize picture and keep my reply short. And Blake hasn't reviewed this comment, so he might disagree with my claims/framing.)

Short answer: Proto-campuses with total sqft below 5000 are very likely too small; and all else being equal, I'd tend to consider 6000+ sqft 'OK/workable', 12,000+ sqft 'quite good', and 20,000+ sqft 'great'.

Longer answer: We can divide indoor spaces into 'space to work in' and 'space to live in', and we can divide MIRI into '~18 researchers' and '~10 non-researchers'. (Some buildings might be a mix of living spaces and working spaces, at least initially.) On that division, there are two very-important 'indoor space' things we want from a (proto-)campus, and two pretty-valuable indoor space things we want (ignoring all the small details like water pressure, lighting, sound isolation...).

  • Very important (#1) - There's enough working space for the ~18 researchers already.

Our Berkeley office was ~10,600 sqft (plus ~1400 sqft CFAR used, and ~2000 sqft for LW and community events), which was a good amount of work space for us. (Somewhat more than we needed at the time, leaving room for staff growth.)

A proto-campus with (e.g.) 5000 sqft would be a tight squeeze no matter how that space is allocated between rooms/buildings, and I'd be surprised if anything less than 5000 sqft were big enough.

(Caveat: what matters is the actual useable space. We've found properties that were listed as <5000 sqft but were more promising than that number alone may make it seem, because there were garages or other nearby buildings that could be turned into work space.)

  • Very important (#2) - We can build or purchase as much working and living space as we're likely to want, and have it be finished within 3 years (or better, 1 year; or better still, we have enough space at the outset).

I'm confident at least some MIRI staff won't live on the campus, even if we have plenty of room. But I'm unsure exactly how many current or future staff will live on the campus, and how many spouses, friends, colleagues, community members, etc. will live or work there. If we moved into a big proto-campus we were excited by, I wouldn't be shocked if as few as 10 or as many as 50 people lived on the campus a couple years later.

I think this would ideally look like a bunch of houses that feel roomy and don't force people to live with more people than they want. (Less Bay Area 'cram people into every spare closet', more 'you and a friend can split a house if you want to'.)

(Hence my note in the OP that we probably want at least 20 acres, and ideally 50+ acres; and that we need to be able to build or otherwise expand.)

  • Pretty valuable (#3) - There's enough living space for at least a few staff, a 0-5 minute walk from working space.

Whereas #2 notes that we want to eventually have lots of living and work space, #3 notes it would be very nice if we initially have at least some living space. If there's enough square footage at the outset that (e.g.) five staff can live there, in addition to the work space, then that's a big plus.

  • Pretty valuable (#4) - There's already enough working space for the ~10 non-researchers too.

This is another very big advantage, but not totally indispensable.

Thinking out loud: Maybe the ideal would be to start with something like ~20,000 sqft of work+community space spread across 1-3 buildings, plus 10-20 medium-sized houses for people to live in, with plenty of capacity to add more work and living spaces as desired? (I haven't been very involved in MIRI's location-choosing process, so this is way-overly-concrete speculation on my part.)

The best places we've seen so far have had more like 1-8 buildings, totaling 7000-25,000 sqft. Most 20+ acre properties we've seen have exactly one building on them, often a 5k+ sqft house.

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-11T22:34:02.551Z · LW · GW

I poked Alex about this, and he didn't have links to specific properties on hand, but he did rattle off some common reasons candidate campuses haven't worked out:

  1. Too far away from anything else.
  2. Related: lack of Uber / UberEats availability.
  3. Reasons to think building more will be very difficult. E.g., the land is in a conservation easement, or it's zoned farmland. (Causing something to no longer be zoned as farmland will generally be impossible or very difficult.)
  4. The buildings don't have enough square footage for us to start researching there in the next few months.
  5. We're not excited enough about the city and/or area.
  6. The land is too small. We've found some really badass buildings, but on a tiny lot surrounded by neighbors, so no space/privacy.

A thing that's an important bonus is if the property is already split into multiple lots. If it's not split into lots, that's another huge preliminary part of the process that has to be done before we develop the property (and it might turn out to be impossible).

Usually you can have one building (and maybe one secondary building) per lot. If you have 50 acres that's designated as a single enormous lot, then you're likely to have to split it before building on it, which adds an additional bureaucratic nightmare to the process.

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-11T22:27:11.387Z · LW · GW

Research Triangle Park is on the list of 14+ cities that Alex classified one tier below our top thirty (as of one month ago): "Moved to backburner; had varying levels of initial interest but also flags; investigated and deprioritized; plausible something could bump them back onto our radar, but seems unlikely." Along with places like Ithaca NY, Salt Lake City UT, Pittsburgh / CMU, Princeton NJ, and Providence RI.

I wasn't involved in the decision process, so I don't know why those fell to the backburner (maybe Alex or Blake will chime in later). Them being classified like that is both a positive update (we thought they were promising enough to seriously consider them) and a negative one (we've already investigated them and ran into obstacles or flags).

I wouldn't be shocked if someplace on that list makes it to our shortlist in the future, and am in favor of hearing more impressions/arguments/considerations about places like those. I mentioned there's maybe a 50/50 chance we change criteria (to pivot away from our focus on finding a campus) soon, in which case I expect the set of promising places to look pretty different. 

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-10T21:59:12.605Z · LW · GW

Thanks! I appreciate the offer, and I'll keep it in mind.

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-10T17:03:07.065Z · LW · GW

I think I like rain and overcast days (though I've never lived in a place like Seattle or London, so maybe I'm underestimating how annoying I'd find it).

Separately, I suspect that my productivity takes a dive when there are fewer daylight hours: my brain wants to say 'the day is over' when the sun starts setting, which is not ideal if that's at 5pm.

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-10T16:27:19.357Z · LW · GW

We've spent physical time in Bellingham, Peekskill, NH, and various other options on our list. (And I expect us to continue making more visits as we weigh the options.)

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-10T16:24:48.935Z · LW · GW

I could see arguments like this changing which areas seem best to us, so I encourage more pitches like this. :) I'll see what Alex and Blake think of your case.

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-10T05:27:22.606Z · LW · GW

We've spent some time visiting southern NH recently, and may do so again as part of picking a destination. :) 

Comment by Rob Bensinger (RobbBB) on Peekskill Lyme Incidence · 2021-05-09T20:36:43.754Z · LW · GW

The CDC thinks reported rates are about 1/10 the actual rates:

A recently released estimate based on insurance records suggests that each year approximately 476,000 Americans are diagnosed and treated for Lyme disease.1,2 This number is likely an over-estimate of actual infections because patients are sometimes treated presumptively in medical practice.

  1. Schwartz AM, Kugeler KJ, Nelson CA, et al. Evaluation of commercial insurance claims as an annual data source for Lyme disease diagnoses. Emerg Infect Dis. 2021;27(2).
  2. Kugeler KJ, Schwartz AM, Delorey M, et al. Estimating the frequency of Lyme disease diagnoses —United States, 2010-2018. Emerg Infect Dis. 2021;27(2).

TickCheck's estimates are based on the CDC's 'multiply by 10x' heuristic, I think? https://www.tickcheck.com/stats/county/new-york/westchester-county/lyme

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-09T18:45:03.549Z · LW · GW

This is basically all the places I mentioned except the Bellingham and Peekskill areas:

  • West Coast: Bend, OR; Eugene, OR; Issaquah, WA; North Bend, WA
  • Rocky Mountain: Boulder, CO; Fort Collins, CO; Bozeman, MT; Missoula, MT
  • Southwest: Austin, TX; Reno, NV
  • Great Plains:
  • Midwest: Urbana-Champaign, IL; Bloomington, IN; West Lafayette, IN; Ann Arbor, MI; Madison, WI
  • South: Asheville, NC; Greensboro, NC; Blacksburg, VA
  • Mid-Atlantic: Rochester, NY; the Philadelphia, PA area (e.g., Lambertville, NJ)
  • New England: the Amherst/Springfield, MA area; the outskirts of the Boston metropolitan area (e.g., Norwood, MA); Portland, ME; the parts of NH nearest Boston; Keene, NH; Portsmouth, NH; Burlington, VT

(Marking our favorite no-property-found-yet options with underline.)

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-09T06:35:31.191Z · LW · GW

Correction from Blake: "that thing you said about [the wave having to get through the strait, etc.], maybe that’s kinda true but not the main thing, the main thing is people probably shouldn’t live right on the coast at sealevel--if you do you need to be ready to get out somehow if the siren sounds, which maybe is hard? But the property is way above sea level and some miles inland.

"The thing about the strait and islands might be true but not very relevant, and [might] be wrong. I think it was my conjecture at the time and I haven’t checked it."

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-09T05:34:08.997Z · LW · GW

Thanks, this is excellent food for thought. I may come back and ask you for more of your models later.

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-09T05:30:50.169Z · LW · GW

"The city proper" meaning Bellingham / Peekskill.

If we moved to Bellingham, I (speculatively) imagine MIRI organizing trips to Seattle or Vancouver once every week or two, including trips to the big universities in those cities, including big-university meetups once or twice a year. I haven't heard discussion of how much rationalists would personally want to hop back and forth between the cities, and I haven't heard a MIRI employee say they'd prefer to live in Seattle and commute. Having to regularly commute from Seattle to Bellingham sounds doable but pretty unpleasant to me. (Maybe better if you're working weird hours, so you can avoid the worst traffic.)

If we moved to Peekskill, I imagine more interaction than that with NYC. (Partly because NYC has more attractions than Seattle/Vancouver; partly because Peekskill has fewer attractions than Bellingham; and partly because the regular trains make it so much more convenient to travel between Peekskill and NYC.)

I can more easily imagine worlds where some MIRI staff lived and worked in NYC itself, though I think MIRI's first-pass goal would be to have as many staff as possible working in the Peekskill area.

Alternatively would working from home some days of the week be viable for many employees?

I already do MIRI work from home a lot in Berkeley. (Well, I did pre-COVID; my living arrangement is weird now.) I think MIRI is pretty pragmatic and case-specific about this, rather than having top-down rules. (Though all else equal, having people in the same place where they can readily interact face-to-face seems better to me.)

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-09T03:32:57.962Z · LW · GW

(Or maybe it would be better to focus on your own preferences and beliefs, since trying to simulate MIRI's might add more noise than signal? Maybe if you're explicit about the delta between your view and the simulated-MIRI view it will be fine.)

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-09T03:30:44.134Z · LW · GW

Our quick and low-confidence conclusion was: something in the neighborhood of 3% chance of major earthquake every ten years; not a major tsunami risk to Bellingham because we aren't directly exposed to the Pacific. The wave has to travel through the de Fuca strait, after which it will spread out, then move between the islands in front of Bellingham in the middle of the sound before reaching Bellingham.

This was a quick and low-confidence conclusion, so I expect our conclusion can be improved on (possibly massively) by someone who has more background knowledge or spends more time on the problem. This was just a first-pass look.

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-09T03:22:31.152Z · LW · GW

I highly appreciate the level of detail here. Breakdowns of the distribution of long-term impressions of Austin weather are great -- a lot more of an update than a single person's take.

I do have to worry that there's a selection effect! People aren't randomly assigned to Austin and aren't forced to stay, so the people who stay (or who are even willing to consider Austin as an option) will skew toward being heat-tolerant.

"Better grocery stores than other places I've lived" - Where have you lived, if you don't mind my asking?

"If you're looking for places near Austin, but out of the city" - If we moved to Austin, I predict the optimal set-up will have some of MIRI in the city and some outside the city, with one of those groups commuting. But it may be hard to achieve that mostly-optimal-for-us set-up.

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-09T03:05:48.248Z · LW · GW

I'm sure we looked at all those areas to some degree, though I wasn't involved in the process that sorted them to 'not top 30', so I don't know what made them look less attractive.

Ithaca specifically is on a list of 14+ cities that Alex classified as one tier below our top thirty: "Moved to backburner; had varying levels of initial interest but also flags; investigated and deprioritized; plausible something could bump them back onto our radar, but seems unlikely."

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-09T02:59:11.908Z · LW · GW

Good question! Blake estimated tsunami risk for Bellingham a few days ago; he's on vacation now, but I've passed this paper to him to see if it's new info to him. Thoughts about this kind of thing (eg, attempts to calculate micromorts) are useful here. It will be a surprising update to me if tsunami risk ends up looking high enough to rule out Bellingham.

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-09T02:48:05.854Z · LW · GW

Thanks, this is helpful stuff. :) I encourage folks to try to gesture at the considerations that seem strongest to them (or that would seem strongest to them if they had slightly more MIRI-ish goals or beliefs, possibly), even if the end result sounds "subjective" or is hard to put into words.

I think Peekskill has ticks and mosquitoes (while Bellingham has neither). If I'm wrong, and the Peekskill area has very mild mosquito seasons, that would be an important positive update for me.

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-09T01:43:09.425Z · LW · GW

One of my co-workers estimated the cost of 'book a hotel room every Saturday night in Manhattan and thus effectively live there on weekends' at $1,000 / month.

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-09T01:25:27.216Z · LW · GW

Yep, 'work outside the city and have fun inside it' is my model, and is part of why I'd expect the train ride to be nice. E.g., I could imagine getting a hotel room and spending Saturdays in the city.

Comment by Rob Bensinger (RobbBB) on MIRI location optimization (and related topics) discussion · 2021-05-09T00:41:09.006Z · LW · GW

'Below 1%' is a very small number!

Comment by Rob Bensinger (RobbBB) on The rationalist community's location problem · 2021-05-08T23:58:52.048Z · LW · GW

Update: https://www.lesswrong.com/posts/SgszmZwrDHwG3qurr/miri-location-optimization-and-related-topics-discussion 

Comment by Rob Bensinger (RobbBB) on Your Dog is Even Smarter Than You Think · 2021-05-05T01:47:14.736Z · LW · GW

It sounds to me like you're thinking of curated as 'vetted' or 'confident', while Ray is thinking of it more as 'representing a direction we want to see LW move in more' (including 'this topic seems neglected here, I'd like to see it get more attention').

It's the difference between 'curation is like publishing a non-replication-crisis-y journal article or encyclopedia article, a durable summary of humanity's knowledge we can call back to and base further conclusions on', versus 'curation is like sending out an email to a private researcher mailing list you run "hey, I'd love to see more discussion here in the vein of X"'.

Comment by Rob Bensinger (RobbBB) on Your Dog is Even Smarter Than You Think · 2021-05-02T15:32:42.446Z · LW · GW

I found the first conversations(?) here very interesting:

It seems like a lot of the words could be swapped for each other or reordered and still look meaningful, so you could get a lot of the way to these conversations(?) just by learning what general categories of words tend to produce surprise/confusion (seemingly irrelevant words like 'stranger', 'friend', 'oops', 'dog', 'hi') versus happiness/engagement ('walk', 'outside', 'beach', 'now', 'soon', 'please', 'want', 'go', 'why', 'when') when you want to go for a walk. I'll certainly be interested to see what falls out of a larger-scale analysis of videos/transcripts.

Comment by Rob Bensinger (RobbBB) on AMA: Paul Christiano, alignment researcher · 2021-05-01T21:18:06.567Z · LW · GW

Source for the blacksmith analogy: I Still Don't Get Foom