Kediska Qaabaynta Abaalmarinta

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

1. Waa maxay wax-soo-saarka aasaasiga ah ee moodeelka abaalgudka ee lammaanaha jawaab-celinta degdegga ah ee la bixiyay?
   • Hal dhibco ah
   • Jawaab qoraal ah oo buuxa
   • Qaybinta itimaalka ee erayada
   • Liiska calaamadaha la door biday
2. What feedback format did the InstructGPT paper collect to train its reward model?
   • A guaranteed correctness label for every possible answer
   • Human rankings of several model responses, converted into pairwise preferences
   • A single sentiment score for each training prompt
   • Only binary safety labels with no response comparisons
3. Waa maxay heerka abaal-marinta-qaabka luminta ay hagaajiso?
   • Qiimaha abaalgudka saxda ah ee jawaab kasta
   • Saxnaanta saadaasha calaamada xigta
   • In jawaabta la doortay ay ka sarreyso tii la diiday
   • Dhererka jawaabaha
4. Waa maxay jabsiga abaal-marinta?
   • Xatooyada qaabka abaalmarinta tartame
   • Habka loo dedejiyo tababarka
   • Sirta calaamadda abaal-marinta
   • Siyaasadda ka faa'iidaysanaysa cilladaha ku jira qaabka abaalgudka si loo helo buundooyin sare iyada oo aan lahayn tayada dhabta ah
5. Sidee qaabka abaalgudka qaab dhismeed ahaan looga soo qaatay qaabka luqadda?
   • Adigoo ku beddelaya madaxa saadaasha calaamada hal madax wax soo saar leh
   • Adigoo ku daraya lakabyo feejignaan badan
   • Adiga oo ka saaraya dhammaan lakabyada beddelka
   • Adigoo labanlaabaya ereyada
6. Sidee buundada abaal-marinta loo istcimaalaa wakhtiga qiimaynta ee muunada-na ugu wanaagsan?
   • Tababbarka N abaal-marinta moodooyinka barbar socda
   • Soo saarista jawaabo badan iyo adeegsiga qaabka abaal-marinta si aad u doorato midka ugu dhibcaha badan
   • Doorashada N calaamadaha ugu caansan
   • Celceliska isbaarooyinka nooca N
7. Why might an RLHF study use an ensemble of reward models during training?
   • To guarantee that reward hacking is impossible
   • To replace the human preference data with exact correctness labels
   • To estimate disagreement or uncertainty across learned preference scorers
   • To ensure every candidate receives the same score
8. Why should reward-model evaluations include held-out comparisons?
   • To ensure training comparisons are never used
   • To make every output identical
   • To guarantee the reward is unbiased
   • To test whether preferences generalize beyond the training comparisons