Some rationality tweets

Will New­some has sug­gested that I re­post my tweets to LessWrong. With some trep­i­da­tion, and af­ter go­ing through my tweets and cat­e­go­riz­ing them, I picked the ones that seemed the most ra­tio­nal­ity-ori­ented. I held some in re­serve to keep the post short; those could be posted later in a sep­a­rate post or in the com­ments here. I’d be happy to ex­pand on any­thing here that re­quires clar­ity.


  1. Test your hy­poth­e­sis on sim­ple cases.

  2. Form­ing your own opinion is no more nec­es­sary than build­ing your own fur­ni­ture.

  3. The map is not the ter­ri­tory.

  4. Thoughts about use­less things are not nec­es­sar­ily use­less thoughts.

  5. One of the suc­cesses of the En­light­en­ment is the dis­tinc­tion be­tween be­liefs and prefer­ences.

  6. One of the failures of the En­light­en­ment is the failure to dis­t­in­guish whether this dis­tinc­tion is a be­lief or a prefer­ence.

  7. Not all en­tities com­ply with at­tempts to rea­son for­mally about them. For in­stance, a hu­man who feels in­sulted may bite you.

Group Epistemology

  1. The best peo­ple en­ter fields that ac­cu­rately mea­sure their qual­ity. Fields that mea­sure qual­ity poorly at­tract low qual­ity.

  2. It is not un­vir­tu­ous to say that a set is nonempty with­out hav­ing any mem­bers of the set in mind.

  3. If one per­son makes mul­ti­ple claims, this in­tro­duces a pos­i­tive cor­re­la­tion be­tween the claims.

  4. We seek a model of re­al­ity that is ac­cu­rate even at the ex­pense of flat­tery.

  5. It is no kind­ness to call some­one a ra­tio­nal­ist when they are not.

  6. Au­mann-in­spired agree­ment prac­tices may be cargo cult Bayesi­anism.

  7. God­win’s Law is not re­ally one of the rules of in­fer­ence.

  8. Science be­fore the mid-20th cen­tury was too small to look like a tar­get.

  9. If schol­ars fail to no­tice the com­mon sources of their in­duc­tive bi­ases, bias will ac­cu­mu­late when they talk to each other.

  10. Some fields, e.g. be­hav­iorism, ad­dress this prob­lem by iden­ti­fy­ing sources of in­duc­tive bias and for­bid­ding their use.

  11. Some fields avoid the ac­cu­mu­la­tion of bias by un­crit­i­cally ac­cept­ing the bi­ases of the founder. Ad­her­ents rea­son from there.

  12. If think­ing about in­ter­est­ing things is ad­dic­tive, then there’s a pres­sure to ig­nore the ex­is­tence of in­ter­est­ing things.

  13. Growth in a sci­en­tific field brings with it in­su­lar­ity, be­cause in­ter­nal progress mea­sures scale faster than ex­ter­nal mea­sures.


  1. It’s re­ally worth­while to set up a good study en­vi­ron­ment. Table, chair, quiet, no com­put­ers.

  2. In emer­gen­cies, it may be nec­es­sary for oth­ers to forcibly ac­cel­er­ate your learn­ing.

  3. There’s a differ­ence be­tween learn­ing a skill and learn­ing a skill while re­main­ing hu­man. You need to de­cide which you want.

  4. It is bet­ter to hold the sword loosely than tightly. This prin­ci­ple also ap­plies to the mind.

  5. Skills are pack­aged into dis­ci­plines be­cause of cor­re­lated sup­ply and cor­re­lated de­mand.

  6. Have a high dis­count rate for learn­ing and a low dis­count rate for know­ing.

  7. “What would so-and-so do?” means “try us­ing some of so-and-so’s heuris­tics that you don’t en­dorse in gen­eral.”

  8. Train hard and im­prove your skills, or stop train­ing and for­get your skills. Train­ing just enough to main­tain your level is the worst idea.

  9. Gain­ing knowl­edge is al­most always good, but one must be wary of learn­ing skills.

In­stru­men­tal Rationality

  1. As soon as you no­tice a pat­tern in your work, au­to­mate it. I sped up my book-writ­ing with code I should’ve writ­ten weeks ago.

  2. Your past and fu­ture de­ci­sions are part of your en­vi­ron­ment.

  3. Op­ti­miza­tion by proxy is worse than op­ti­miza­tion for your true goal, but usu­ally bet­ter than no op­ti­miza­tion.

  4. Some tasks are costly to re­sume be­cause of men­tal mode switch­ing. Max­i­mize the cost of ex­it­ing these tasks.

  5. Other tasks are easy to re­sume. Min­i­mize ex­ter­nal costs of re­sum­ing these tasks, e.g. by leav­ing soft­ware run­ning.

  6. First eat the low-hang­ing fruit. Then eat all of the fruit. Then eat the tree.

  7. Who are the mas­ters of for­get­ting? Can we learn to for­get quickly and de­liber­ately? Can we just for­get our vices?

  8. What sorts of cul­tures will en­dorse causal de­ci­sion the­ory?

  9. Big agents can be more co­her­ent than small agents, be­cause they have more re­sources to spend on co­her­ence.