# Saving the world in 80 days: Epilogue

80 days ago, I started a pro­duc­tivity sprint for be­ing use­ful in the field of AI al­ign­ment. My main goal was to have a bet­ter self-model in or­der to push my­self with­out burn­ing out. I’ll sep­a­rate this into knowl­edge, emo­tions, health, and notic­ing.

## Knowledge

I’ve felt my mind ex­pand more in these past 80 days than any semester in col­lege due to read­ing these dense, no-fluff text­books. How To Prove It, Jayne’s Prob­a­bil­ity The­ory, & Tao’s “Anal­y­sis I” have grown me a LOT (I read 4 chap­ters of Jayne’s as recom­mended in Miri’s re­search guide, and I’m cur­rently on ch. 5 on Tao’s).

I’ve no­tice a ten­dency in my­self to get im­pa­tient and de­sire to skip the ex­er­cises, and that screwed me over when try­ing to get through Lin­ear Alge­bra Done Right where I only made it to ch. 4. So, as an ex­er­cise for the reader, don’t skip the ex­er­cises for the reader, or you’ll be weak sauce.

## Emotions

Every­thing was go­ing great un­til 40 days ago: I lost a friend, had to fire a friend, moved, de­con­verted from Chris­ti­an­ity, ex­pe­rienced ex­treme ro­man­tic re­la­tion­ship un­cer­tainty, and just felt lonely. I’ve read and re-read Valen­tine’s griev­ing well, and Squir­re­lin­hell’s ex­pla­na­tion of Gendlin’s fo­cus­ing and more gen­eral emo­tional tun­ing. They’ve helped a lot. I’m definitely still deal­ing with feel­ing lonely, but I feel like I’m mostly through the rest which is re­as­sur­ing.

I’ve no­ticed the ten­dency to dis­tract my­self away from the pain with fic­tion, but the proper re­sponse is to look into the pain, ac­knowl­edg­ing that it ex­ists, that it’s part of re­al­ity, that it’s still true even if I dis­tract my­self from it.

## Health

I used to eat out a lot and get pay­days (a candy bar) from the vend­ing ma­chine. I no­ticed I’d get a pay­day when work would get tough. It was a form of pica for me like read­ing fic­tion is now. ~50 days ago, I started eat­ing meal squares for lunch, and eat­ing an uber large, very fatty salad for din­ner. I’ve felt very sa­ti­ated, and I’ve saved money and time de­spite eat­ing bet­ter qual­ity food and prepar­ing din­ner. Plus! I can make a mean salad that leaves oth­ers green with envy. (Did you read 1 pun, or 2?)

I’ve also no­ticed that I drink more and more coffee over a cou­ple weeks un­til I’m an anx­ious mess, quit coffee, wait a week, rinse, re­peat. Now I drink ~6 cups of de­caf green tea in­stead which fixed that prob­lem!

I’ve also, also no­ticed that I have less en­ergy if I haven’t danced in a few days. I usu­ally dance right af­ter work, but some­times work lasts till 6 and that gets skipped. Shift­ing my evening sched­ule might help, but also just leav­ing work at 5 pm would fix it too.

## Noticing

Med­i­tat­ing re­ally boot­strapped my abil­ity to no­tice my im­me­di­ate re­ac­tions such as flinch­ing away from pain. I’ve mostly done Squir­re­lin­hell’s sense your body with ex­treme clar­ity, and while do­ing that I’d no­tice my back hurt­ing & want­ing to stop med­i­tat­ing. I would just switch my fo­cus to the pain in my back and just watch it. The pain would wa­ver for like 5 sec­onds and then stop. Same with fo­cus­ing on “want­ing to stop med­i­tat­ing”. This skill helped a lot with the emo­tional sec­tion, and I’m glad I med­i­tated.

# Fu­ture work

Study­ing & med­i­tat­ing will con­tinue to be a fo­cus for me, and I’m go­ing to spend this week­end work­ing out feel­ing lonely. I’m go­ing to go through An­drew Ng’s ma­chine learn­ing course for a week to see if it’s a low spoons ac­tivity for me (I very much en­joy pro­gram­ming, so I pre­dict it will be).

[1] The 80 days is over, and yes, I know, I didn’t save the world (darn!). It was mainly a play on “Around the world in 80 days”, lol.

No nominations.
No reviews.
• Con­grats on see­ing this through, and I’m im­pressed with your at­ten­tion to both in­tel­lec­tual progress as well as emo­tional well be­ing.

The 40 day mark sounds ex­tremely rough, that’s a lot to have dealt with at once. But again, sounds like you’re at least pro­cess­ing and ap­proach­ing it in a fairly sane, healthy way.

• I ap­pre­ci­ate you ex­plic­itly stat­ing that it was a lot to deal with. I was ac­tu­ally a bit em­bar­rassed be­cause I didn’t get as much done as I ex­pected, so, again, thanks Rae­mon.

• Con­grat­u­la­tions on finish­ing this 80 days jour­ney.

It is always nice to read some feed­back from peo­ple af­ter they finished a long pro­ject like that.

I have been read­ing some of Nate Soares’ post re­cently, and I’m still fight­ing with my­self to start try­ing to do some­thing similar. I hope I can start some­thing like that soon. Your post gave me an ex­tra mo­ti­va­tion to do it.

Can you de­tail a lit­tle more how much time you spent on each ac­tivity you set on the other post? If you didn’t fol­lowed the set amount, when did you de­cide to change them?

About the study/​read­ing ses­sions, how much effi­ciency do you es­ti­mate you could achieve? By effi­ciency I mean the time you were effec­tively do­ing that, not con­sid­er­ing some breaks or in­ter­nal strug­gles, com­pared with the al­lot­ted time .

• Thanks re­nato!

## Re­gard­ing your first set of ques­tions:

Read­ing: origi­nally 3 hr. This changed to 0-3 hours de­pend­ing on when I woke up in the morn­ings, which meant go­ing to bed around 9-10, which meant mak­ing a habit of try­ing to fall asleep when I get in bed. I did try to read in the evenings as well, but my eyes would glaze over af­ter work­ing for the day.

Ten­sor flow: origi­nally 2 hr. I dropped this af­ter the first week due to work tak­ing up 2.2x as much time as I ini­tially pre­dicted. I also felt like I wasn’t ac­tu­ally learn­ing any­thing while go­ing through Google’s tu­to­ri­als, and TurnTrout con­vinced me to just learn the pre-reqs and the­ory of ML first.

AI-Safety read­ing: origi­nally 1 hr. This be­came 1-3 hours de­pend­ing on how in­ter­ested I was in. I dropped this though af­ter I moved 40 days ago due to habit changes and then sim­ply for­get­ting! I re­ally en­joyed it and it was a low-spoons ac­tivity for me.

Med­i­tate: origi­nally 1 hr. This be­came 0-1 hours de­pend­ing on the day. I ex­per­i­mented with do­ing it at differ­ent times, but dur­ing my lunch break is prob­a­bly the best for me at the mo­ment.

Week­ends are a differ­ent an­i­mal. If I had a free one & wasn’t ex­pe­rienc­ing emo­tional prob­lems, I would tear into a book, med­i­tate a lot, read AI pa­pers, and just get a lot of read­ing on LW and SlateS­tarCodex done which was great! I re­ally wish I didn’t have to work so that could be ev­ery day.

So to an­swer your ques­tion, I changed what I was do­ing af­ter giv­ing it a solid try and ad­just­ing from there if I needed to. After do­ing that for a cou­ple of months, I have a much bet­ter idea of how to do these types of things now.

## Re­gard­ing “How much effi­ciency dur­ing read­ing ses­sions?”

I in­ter­pret that to mean “how many pages per day”/​”How many chap­ters per week”/​”How many books per month”. If that’s cor­rect, then I would say I could (right now) learn a sub­ject/​book a month. Like I could read Lin­ear Alge­bra Done Right in less than a month and Tao’s Anal­y­sis I & II in less than two months, while do­ing all of the ex­er­cises.

If I didn’t have to work, I would pre­dict that I could half that time and finish one in less than 2 weeks.

• I in­ter­pret that to mean “how many pages per day”/​”How many chap­ters per week”/​”How many books per month”. If that’s cor­rect, then I would say I could (right now) learn a sub­ject/​book a month. Like I could read Lin­ear Alge­bra Done Right in less than a month and Tao’s Anal­y­sis I & II in less than two months, while do­ing all of the ex­er­cises.

It wasn’t ex­actly that, but it still a very good in­for­ma­tion.

I will try to de­scribe it with my own ex­pe­rience.

When I read Nate’s post, he said that he stud­ied 5~7 hours per day. At that time, I was log­ging ex­actly how long I was study­ing each day to mea­sure how much time it took me to do some tasks re­lated to study­ing to es­ti­mate how long the whole pro­cess it would take. In my case, I started the clock, and started read­ing. If I took a break, the clock was stopped.

How­ever, the most pro­duc­tive days I had, when I was in my best hu­mor and ev­ery­thing worked perfectly, I could reach around 6~7 hours. This left me won­der­ing if he was re­ally hu­man to keep such high num­ber for a long pe­riod.

When I checked some­where else (prob­a­bly some com­ments) he said that his study effi­ciency was around 30 % (not sure about the num­ber, but some­thing like that). The rest of the time he spent con­vinc­ing him­self to do it, to con­tinue, or try­ing to pro­cess what he had read.

Be­cause of that, it was an ex­traor­di­nary amount, but it was not an or­der of mag­ni­tude higher than my effort.

I tried to do some more re­laxed mea­sure to com­pare with his num­ber, but I usu­ally took some long breaks to do some­thing else, and I could not con­sider that into a study time.

When I read your post, it made me cu­ri­ous to know if those 3 hours would be com­bined into 3*30 → 90 hours to read a book and learn a sub­ject, or 3*eff*30 → ~30 hours ?

edit: I mis­take the num­bers, the sec­ond one should also have been 30.

• Thanks for the clar­ifi­ca­tion!

I un­der­stand your first equa­tion 3*30= 90 hr (3 hrs/​day* 30 days = 90 hrs to com­plete a book). The sec­ond one is a bit con­fus­ing. 3*eff*90 = ~30 hours. (3 hrs/​day * 19 effi­ciency * 90 ? = ~30 hours) Was the sec­ond 90 sup­posed to be 30 to make effi­ciency 30%?

I definitely think effi­ciency plays a huge role, and I’d say (Hours/​day *days * effi­ciency = to­tal ac­tual hours for a 100% effi­cient per­son) would be my an­swer.

(hope­fully) Re­lated: an over­all good frame to tackle this is fo­cus­ing on in­creas­ing quan­tity and qual­ity of time spent.

In­creas­ing quan­tity:

• Tak­ing time from other ac­tivi­ties and put it in this.

• Grad­u­ally build­ing up to longer and longer times. (6-7 hours on a good day is fan­tas­tic, and I ap­plaud you! So what about on av­er­age? 4-5 hours?)

• Notic­ing why you stop. Does your mind want to quit? Are you phys­i­cally tired or have tired eyes?

In­creas­ing qual­ity:

• Mak­ing sure you have a good text­book.

• Sim­ply read­ing faster w/​o los­ing com­pre­hen­sion (Imag­in­ing hy­po­thet­i­cal “Gun to the head, read as fast as you can w/​o los­ing com­pre­hen­sion for 5 min­utes, you will be quizzed” might be a good ex­er­cise to play around with)

• Not wast­ing time be­tween ac­tivi­ties (like go­ing from one ex­er­cise to the next) but also!:

• Know­ing how to make use of your sub­con­scious by let­ting your mind wan­der (I don’t have a gears-level model of this one yet, but I have in­tu­itions that it’s im­por­tant)

May I ask why you would like to com­pare study paces?

• I un­der­stand your first equa­tion 3*30= 90 hr (3 hrs/​day* 30 days = 90 hrs to com­plete a book). The sec­ond one is a bit con­fus­ing. 3*eff*90 = ~30 hours. (3 hrs/​day * 19 effi­ciency * 90 ? = ~30 hours) Was the sec­ond 90 sup­posed to be 30 to make effi­ciency 30%?

Yes, I put the wrong num­ber.

I would say my av­er­age was 4 hours dur­ing al­most two months. I could no­tice it slowly in­creas­ing, but ev­ery time I need to change to a new sub­ject/​differ­ent kind of ac­tivity (from read­ing a non-math book to one that re­quires more ex­er­cises) my time plum­meted. After a few days I could in­crease it again.

I think it was not tired­ness per se, be­cause I did not have any prob­lem with stop­ping study­ing and start­ing read­ing fic­tion or blogs. I guess it was more that I took too much in­for­ma­tion and needed to left it or­ga­nize it­self into my mind.

I wanted to com­pare the study paces be­cause 3 hours of effec­tive study is much more de­mand­ing than a 3 hours study slot time.

I can eas­ily re­serve 3~6 hours a day for mainly study­ing, if that in­cludes tak­ing some breaks to re­ply emails, or fix some small un­re­lated bugs. I can prob­a­bly di­vide it into differ­ent sub­jects and try them for some time and even­tu­ally drop what does not work.

How­ever, 3~6 hours of effec­tive or pure study per day is some­thing that would make me much more tired. In that case I would try to fo­cus on just one thing now, while try­ing to work on some meta-tech­niques/​skills to fo­cus on long-term.

It is more a plan­ning ques­tion to ad­just how much I should fo­cus on a sub­ject or on a broad range of top­ics/​bugs. But it ac­tu­ally is not that im­por­tant.

• I un­der­stand now! haha

My study time was a non-in­ter­rupted 1-3 hr block. This made it eas­ier to get in the zone and have rele­vant de­tails in my work­ing mem­ory.

Go­ing for longer than that (4-6 hr), I’d pre­dict I would need to take a walk out­side and just think of noth­ing to let my sub­con­scious do it’s thing. I haven’t done that more than once or twice this sum­mer, so I’m not sure what would be nor­mal for me.

TurnTrout has a lot more ex­pe­rience do­ing that than me, and he’d be a great re­source for any of these type ques­tions.

We ac­tu­ally have a dis­cord server with sev­eral peo­ple study­ing miri-re­lated ma­te­ri­als if that’s some­thing that in­ter­ests you.

• Thank you for the in­for­ma­tion.

I’ve fi­nally finished some kind of manda­tory ma­te­rial I had to cover for some ex­ams.

Even as I prob­a­bly failed them, I still learned some stuff in the pro­cess, and now I’m al­most free to pur­sue some­thing more in­ter­est­ing.

Thank you for the in­vi­ta­tion.

Although, the ar­tifi­cial in­tel­li­gence and re­lated top­ics in­ter­est me a lit­tle, I’m still lack­ing sev­eral of the ba­sic req­ui­sites.

I will try to fo­cus on them first, and then de­cide what to do next.

• The most po­tent mo­ti­va­tional blend I have found is a mix of emo­tion­ally up­dat­ing to x-risk be­ing real, and of delight­ful cu­ri­os­ity about dis­cov­er­ing how the math­e­mat­i­cal world ticks. Echo­ing Nate: once you get started, the learn­ing part is re­ally not that hard (in my opinion).

• I used to eat out a lot and get pay­days from the vend­ing ma­chine. I no­ticed I’d get a pay­day when work would get tough.

Note for any­one con­fused by this bit, as I ini­tially was:

A PayDay seems to be a candy bar.

• Edited, thanks!

• Con­grats on your med­i­ta­tion! I re­mem­ber com­ment­ing on your Prologue, about 80 days ago. Time flies!

Good luck with your ML jour­ney. I did the 2011 Ng ML course, that uses Mat­lab, and Ng’s DL spe­cial­iza­tion. If you want to get a good grasp of re­cent ML I would recom­mend you to di­rectly go to the DL spe­cial­iza­tion. Most of the origi­nal course is in the newer course, and the DL spe­cial­iza­tion uses more re­cent libraries (tf, keras, numpy).

• I’ll look into that course be­fore I start, thanks for the recom­men­da­tion!