Kvíz spojený s průvodcem · Střední úroveň

Jak si umělá inteligence vede v kvízu advokátní zkoušky

Testuje strukturu zkoušky Uniform Bar Exam, kritiku tvrzení GPT-4 na 90. percentilu, a proč kontaminace a volby hodnocení ovlivňují skóre AI.

Související vodicí cestyAI a advokátní zkouška
Otázka 1 z 8

Which Uniform Bar Exam component has 200 multiple-choice questions and counts for half the total score?

Náhled každé otázky8 otázky

Odpovědi a vysvětlení se zobrazují, jak budete v kvízu.

  1. Which Uniform Bar Exam component has 200 multiple-choice questions and counts for half the total score?
    • The Multistate Performance Test (MPT)
    • The Multistate Essay Examination (MEE)
    • The Multistate Bar Examination (MBE)
    • The Multistate Professional Responsibility Examination (MPRE)
  2. What simulated Uniform Bar Exam score did OpenAI's GPT-4 technical report cite in 2023?
    • About 298 out of 400
    • About 250 out of 400
    • About 350 out of 400
    • About 198 out of 400
  3. According to Eric Martínez's 2024 critique, why was GPT-4's 90th-percentile figure misleading?
    • The model was shown the answer key before testing
    • It was compared with February test takers, a group that includes many repeat takers
    • Only the performance test section was used
    • The comparison group was law professors, not applicants
  4. Who graded GPT-4's essay answers in the 2023 simulated bar exam, according to the critique described in the guide?
    • Official NCBE-trained bar graders
    • A committee of a state bar
    • The researchers themselves, using published sample answers
    • A second AI model working as an automatic grader
  5. What did Bommarito and Katz find when they tested GPT-3.5 on MBE practice questions in late 2022?
    • It answered every question correctly
    • It refused to answer legal questions
    • It scored below random chance
    • It answered about half correctly: above chance, below passing
  6. Why is training-data contamination a specific concern for AI bar exam results?
    • Past bar questions circulate online and may have been in the model's training data
    • Bar exams are only given in printed form, so models cannot read them
    • The NCBE deliberately publishes wrong answers
    • Models are forbidden from seeing legal text during training
  7. Why do researchers simulating a bar exam score have to approximate the conversion from raw to scaled scores?
    • Scaled scores are chosen at random each year
    • Models produce answers that cannot be scored
    • Each jurisdiction uses a completely different set of questions
    • The NCBE's full equating process is not public
  8. What does the guide say the NCBE's NextGen bar exam gives more weight to?
    • Memorizing a longer list of rules
    • Integrated skills such as research and drafting
    • Speed on multiple-choice questions
    • Oral argument before a panel of judges