ManySpace
English Giriş yap Kayıt ol

Yapay ZekâUzmanlıkAIML-RSP-012

Scalable Oversight and Dangerous-Capability Evaluation

Supervising systems whose outputs humans struggle to check, and evaluating for high-risk capabilities before deployment.

Bu konunun henüz Türkçe çevirisi yok, bu yüzden İngilizce gösteriliyor.

Zorluk
5 / 5
Önem
3 / 5
Çalışma süresi
1–2 sa
Çalışma türü
Karma

Neden önemli

The learner can explain why human evaluation breaks down for superhuman outputs and describe one scalable-oversight proposal and one dangerous-capability evaluation.

Bu konuya giden rota

Rota AIML-RSP-012
What Is a Model? Learning from Data Tasks: Regression, Classification, and Beyond Features and Targets Datasets and Splits Classification Framing Linear Regression I: Prediction and Residuals Tokens and Tokenization Sampling and Splits Train, Validation, and Test Errors Scores and Logits Loss and Fitting: Squared Error Distribution Shift Offline Evaluation Multiclass Scores and Softmax Autoregression Objective Functions and Empirical Risk Next-Token Prediction Gradient Descent in Machine Learning Likelihood Training vs Serving Generation Metrics Stochastic and Mini-batch Optimization Monitoring and Drift Cross-Entropy Loss The Language-Modeling Objective Human Oversight Pretraining LLM Evaluation Scalable Oversight and Dangerous-Capability Evaluation