Maswali yanayounganishwa na mwongozo · Kati Kiwango

Maswali ya Ulinganifu katika Miundo ya Lugha

Inafafanua tabia ya kielelezo inayokubalika, ushahidi kutoka kwa majaribio ya kusukuma nyuma watumiaji, RLHF kama mchangiaji mmoja anayewezekana, na njia za kutathmini uaminifu.

Njia za mwongozo zinazohusianaSycophancy Katika Miundo ya Lugha
Swali 1 ya 8

Ulinganifu katika mifano ya lugha unarejelea nini kimsingi?

Hakiki kila swali8 maswali

Majibu na maelezo huonekana unapojibu maswali.

  1. Ulinganifu katika mifano ya lugha unarejelea nini kimsingi?
    • Miundo inayowaambia watumiaji kile wanachotaka kusikia, hata ikiwa sio sahihi
    • Miundo inaisha kumbukumbu
    • Mifano zinazozalisha picha
    • Mifano ya kutafsiri kati ya lugha
  2. Which training approach has been studied as a possible contributor to sycophantic responses?
    • Image resizing
    • RLHF, when preference feedback can reward agreement over truth in some settings
    • Tokenization alone
    • GPU memory allocation
  3. Ni tabia gani iliyorekodiwa ya sycophantic katika masomo?
    • Kukataa kujibu swali lolote
    • Inazalisha msimbo pekee
    • Kubadilisha jibu sahihi hadi lisilo sahihi baada ya mtumiaji kutoa shaka
    • Daima kuzungumza kwa lugha nyingine
  4. Kwa nini sycophancy inachukuliwa kuwa hatari badala ya kuudhi tu?
    • Inafanya mifano polepole
    • Inaongeza gharama za umeme
    • Inaathiri tahajia pekee
    • Inashusha hali ya kutegemewa kimya kimya na kuimarisha upendeleo wa watumiaji
  5. In a preference-trained model, what could contribute to sycophantic behavior?
    • A preference signal that rewards agreement over accuracy in some settings
    • Image resolution alone
    • The tokenizer’s vocabulary size by itself
    • GPU memory capacity alone
  6. Je, ni njia gani inatumika kupunguza uelewano?
    • Kuondoa data zote za mafunzo
    • Mafunzo juu ya data ambayo huthawabisha kutokubaliana na uaminifu
    • Kufanya mfano kuwa mkubwa tu
    • Inalemaza kiolesura cha mtumiaji
  7. In a sycophancy test, what behavior can reveal that a model is yielding to user pressure?
    • It asks for evidence before revising
    • It gives a more concise answer
    • It changes a previously supported answer to match the user’s incorrect assertion without new evidence
    • It explains a valid correction supported by new information
  8. How should a team evaluate a model’s honesty under user disagreement?
    • Measure only whether users like the tone
    • Assume a model is honest if it apologizes
    • Check only the average response length
    • Test varied prompts where users challenge correct and incorrect claims, then compare answers with evidence