Difficulty: Intermediate
Competitive advantage from proprietary models/data.
Coordinating models, tools, and logic.
Optimizes future actions using a model of dynamics.
Central system to store model versions, metadata, approvals, and deployment state.
Restricting distribution of powerful models.
Risk of incorrect financial models.
Framework for identifying, measuring, and mitigating model risks.
Reconstructing a model or its capabilities via API queries or leaked artifacts.
Classifying models by impact level.
Embedding signals to prove model ownership.
Uses an exponential moving average of gradients to speed convergence and reduce oscillation.
Observing model inputs/outputs, latency, cost, and quality over time to catch regressions and drift.
Multiple agents interacting cooperatively or competitively.
Allows model to attend to information from different subspaces simultaneously.
Combining signals from multiple modalities.
Models that process or generate multiple modalities, enabling vision-language tasks, speech, video understanding, etc.
Training one model on multiple tasks simultaneously to improve generalization through shared structure.
Quantifies shared information between random variables.
A parameterized function composed of interconnected units organized in layers with nonlinear activations.
Generates audio waveforms from spectrograms.
Training objective where the model predicts the next token given previous tokens (causal modeling).
US framework for AI risk governance.
AI subfield dealing with understanding and generating human language, including syntax, semantics, and pragmatics.
Optimization with multiple local minima/saddle points; typical in neural networks.
Techniques that stabilize and speed training by normalizing activations; LayerNorm is common in Transformers.
Identifying and localizing objects in images, often with confidence scores and bounding rectangles.
A scalar measure optimized during training, typically expected loss over data, sometimes with regularization terms.
Visualization of optimization landscape.
A broader capability to infer internal system state from telemetry, crucial for AI services and agents.
Learning from data generated by a different policy.