# AI Safety Prerequisites Course: Revamp and New Lessons

Nine months ago we, RAISE, have started cre­at­ing a Math Pr­ereq­ui­sites for AI Safety on­line course. It has mostly MIRI re­search re­lated sub­jects: set the­ory, com­putabil­ity the­ory, and logic, but we want to add ma­chine learn­ing re­lated sub­jects in the fu­ture. For 4 months we’ve been adding new les­sons and an­nounc­ing them on LessWrong. Then we stopped, looked back and de­cided to im­prove their us­abil­ity. That’s what we’ve been busy with since Au­gust.

## News since the last post

1. Big up­date of 7 lev­els we had pre­vi­ously pub­lished, which you can see in the pic­ture above. The les­sons use text­books, which you will need to fol­low along. Pre­vi­ously les­sons looked like “read that sec­tion; now solve prob­lems 1.2, 1.3, 1.4c from the text­book; now solve these ad­di­tional prob­lems we came up with”. Now our les­sons still say “read that sec­tion”, but the prob­lems (and their solu­tions, in con­trast to many text­books, which don’t provide solu­tions) are in­cluded in les­sons them­selves. Ad­di­tional prob­lems are now op­tional, and we recom­mend that stu­dents skip them by de­fault and do them only if they need more prac­tice. New lev­els in Logic, Set The­ory, and Com­putabil­ity tracks will be like that as well.

2. Level 1 was very long, con­sisted of 45 pages of read­ing, and could take 10 hours for some­one un­fa­mil­iar with logic. We sep­a­rated it into smaller parts.

3. Two new lev­els. Level 8.1: Proof by In­duc­tion. Level 8.2: Aba­cus Com­putabil­ity.

If you study us­ing our course, please give us feed­back. Leave a com­ment here or email us at raise@aisafety.camp, or through the con­tact form. Do you have an idea about what pre­req­ui­sites are most im­por­tant for AI Safety re­search? Do you know an op­ti­mal way to learn them? Tell us us­ing the same meth­ods or col­lab­o­rate with us.

Can you check if a math­e­mat­i­cal proof is cor­rect? Do you know how to make proofs un­der­stand­able and easy to re­mem­ber? Would you like to help to cre­ate the pre­req­ui­sites course? If yes, con­sider vol­un­teer­ing.

• Seems in­ter­est­ing, thanks!

I definitely think ma­chine learn­ing top­ics are use­ful. Given that there’s so much stuff out there and you can only cover a small frac­tion of it, maybe re­cent ma­chine learn­ing top­ics are a point of com­par­a­tive ad­van­tage, even. The best text­book on set the­ory is prob­a­bly pretty good already.

Another ser­vice that could take ad­van­tage of pre-ex­ist­ing text­books is short sum­maries, de­signed to give peo­ple just enough of a taste to make an in­formed de­ci­sion about read­ing said good text­book. Prob­a­bly eas­ier than de­vel­op­ing a course on al­gorith­mic in­for­ma­tion the­ory, or cir­cuit com­plex­ity, or what­ever.

• maybe re­cent ma­chine learn­ing top­ics are a point of com­par­a­tive advantage

Do you mean re­cent ML top­ics re­lated to AI safety, or just re­cent ML top­ics?

RAISE is already work­ing on the former, it’s an­other course which we in­ter­nally call “main track”. Right now it has the fol­low­ing um­brella top­ics: In­verse Re­in­force­ment Learn­ing; Iter­ated Distil­la­tion and Am­plifi­ca­tion; Cor­rigi­bil­ity. See https://​​www.aisafety.info/​​on­line-course

• Con­sider not wast­ing your reader’s time with hav­ing to reg­ister on gras­ple to be pre­sented with 34-euro pay­wall.

• Regis­tra­tion and ac­cess to the les­sons is com­pletely free. Where do you see a pay­wall?

• Oh, sorry. Javascript shenani­gans seem to have sent me into an­to­her course, works fine on a clean browser.

• Per­haps he’s talk­ing about text­books?