Idanwo ti o sopọ mọ itọsọna · Lile Ipele

AI Aabo adanwo

Agbara idanwo, titete, ilokulo, ere sipesifikesonu, ẹgbẹ-pupa, abojuto, ati awọn eewu ti o kọja sisẹ iṣelọpọ ati igbero imuṣiṣẹ.

Jẹmọ itọsọna awọn ọnaAI Aabo
Ibeere 1 ti 8

Bawo ni itọsọna ṣe fireemu eewu ailewu AI kọja awọn eto lọwọlọwọ ati ilọsiwaju?

Awotẹlẹ gbogbo ibeere8 ibeere

Awọn idahun ati awọn alaye han bi o ṣe n mu ibeere naa.

  1. Bawo ni itọsọna ṣe fireemu eewu ailewu AI kọja awọn eto lọwọlọwọ ati ilọsiwaju?
    • Lati awọn ipalara faramọ ati ilokulo nipasẹ awọn eewu to lagbara lati awọn eto ti o lagbara pupọ
    • Nikan bi ibeere ti idilọwọ awọn abajade arínifín
    • Nikan gẹgẹbi ọrọ igbẹkẹle ohun elo
    • Gẹgẹbi ọrọ-ọrọ fun cybersecurity nikan
  2. In the guide’s capability-alignment-security mental model, what does capability describe?
    • Whether the system’s goals match intent
    • Whether an adversary can misuse it
    • How many training examples it has
    • What a system can do
  3. What does specification gaming mean in AI safety?
    • Correctly following every broader human value
    • Meeting the literal objective while missing the designer’s intended goal
    • Finding a security flaw in a database
    • Using a smaller model to save compute
  4. What is the risk in relying only on output filters for a tool-using agent?
    • Filters always change model weights
    • A filter prevents all misuse
    • The system may take multi-step actions or be manipulated beyond the filtered text
    • Tool use only occurs during training
  5. Why include red-teaming in a safety program?
    • To improve marketing copy
    • To prove the system has no risks
    • To probe adversarially for harmful behaviors and weaknesses
    • To remove the need for deployment monitoring
  6. What should a safety evaluation examine beyond a polished model card?
    • Dangerous capabilities, deceptive behavior, deployment pressure, and safeguards
    • Only the product’s user interface
    • Only benchmark accuracy
    • Awọn nọmba ti paramita awoṣe
  7. What does instrumental convergence describe in AI safety arguments?
    • All AI systems share one objective
    • Every model is conscious
    • A model’s output tokens become instruments
    • Different goals may create incentives for similar subgoals such as resources or persistence
  8. Why does the guide argue for broader public AI literacy?
    • Only experts are affected by AI safety
    • Governance needs public understanding, not only specialist knowledge
    • Literacy replaces technical evaluation
    • Public understanding guarantees safe deployment