Maswali ya Mashambulizi ya Kuchimba na Kuiba

Majibu na maelezo huonekana unapojibu maswali.

1. Ni wazo gani la msingi nyuma ya shambulio la uchimbaji wa mfano?
   • Kuuliza modeli na kumfundisha mbadala kuiga matokeo yake
   • Kuingilia kwenye seva ili kunakili faili za uzani moja kwa moja
   • Kuhonga mfanyakazi ili kuvuja mkusanyiko wa data wa mafunzo
   • Inapakia API kupita kiasi hadi ivunjike na kutupa kumbukumbu
2. Kwa nini kurudisha vekta za uwezekano hurahisisha uchimbaji kuliko kurudisha lebo ya juu-1 pekee?
   • Inasimba majibu, na kupunguza kasi ya mshambuliaji
   • Inavuja maelezo zaidi kwa kila hoja, kwa hivyo maswali machache yanahitajika
   • Inaficha mpaka wa uamuzi kabisa
   • Inaongeza gharama ya seva ili washambuliaji wakate tamaa
3. Ni mbinu gani ya ulinzi ambayo inapunguza moja kwa moja habari inayovuja kwa kila hoja?
   • Kuongeza GPU zaidi kwenye seva
   • Kuongeza hesabu ya parameta ya mfano
   • Inarejesha matokeo ya mviringo au ya juu-1 pekee
   • Kuchapisha data ya mafunzo kwa uwazi
4. What did the 2024 production-language-model extraction paper report about GPT-3.5 Turbo?
   • It reconstructs every GPT-3.5 parameter exactly for under $20
   • It obtains the model’s training dataset directly
   • It extracts a complete model checkpoint from the API
   • It recovers the embedding-projection layer; the paper estimates under $2,000 in queries for GPT-3.5 Turbo
5. Kwa nini modeli ya mbadala iliyoibiwa ni hatari zaidi ya mapato yaliyopotea tu?
   • Huwasha mashambulizi ya kisanduku cheupe nje ya mtandao kama kuunda mifano ya wapinzani
   • Inaboresha usahihi wa mfano wa asili
   • Inafuta kiotomati muundo asili
   • Inahitajika kurejesha asili
6. In ActiveThief, what does the combined DFAL + K-center query strategy aim to balance?
   • Boundary-proximate informative samples and reduced redundancy among selected samples
   • Exact recovery of all secret model weights and a guaranteed substitute
   • Random sampling only and avoiding all substitute-model predictions
   • Higher API throughput and removal of the query budget
7. What can API rate limits do against a model-extraction attempt?
   • Prove that no substitute model can be trained
   • Remove all information from each permitted prediction
   • Make large-scale querying slower or more expensive
   • Make model outputs impossible to compare
8. What does the DynaMarks paper propose for a model-extraction defense?
   • It adds a logo to every generated answer
   • It encrypts all model weights at rest
   • It guarantees ownership of every copied production model
   • It can embed a signature in a surrogate model by dynamically changing API outputs using secret parameters