Advanced AI Vocabulary
Advanced terminology, precise definitions, and distinctions across ten areas of AI. One standalone assessment.
FORMATSingle answer
QUESTIONS40 of 120
TIME LIMIT60 minutes
PASS MARK85%
What this exam covers
Distinguish 120 terms across architecture, training, inference, retrieval, agents, evaluation, security, operations, learning, and governance.
- Model architecture
- Training & adaptation
- Inference & decoding
- Retrieval & representations
- Agents & integration
- Evaluation & measurement
- Security & privacy
- Serving & operations
- Learning paradigms & data
- Strategy & governance
Before you begin
Select one best answer per question. You can move between questions, change an answer, and flag items for review until submission. Unanswered questions count as incorrect.
The timer continues when you leave the page. Your progress is saved to your account. Submit before the deadline, or your saved answers are scored when time expires.
The vocabulary syllabus
120 terms · 10 areasEach attempt samples four terms from every area. Questions ask you to identify a concept or distinguish its definition.
Model architecture12 terms
- Transformer
- Self-attention
- Cross-attention
- Positional encoding
- Rotary position embedding (RoPE)
- Mixture of experts (MoE)
- Sparse activation
- Grouped-query attention (GQA)
- Latent space
- Autoregressive modeling
- Diffusion model
- Attention mechanism
Training & adaptation12 terms
- Pretraining
- Supervised fine-tuning (SFT)
- Parameter-efficient fine-tuning (PEFT)
- Low-rank adaptation (LoRA)
- QLoRA
- Knowledge distillation
- Catastrophic forgetting
- Gradient checkpointing
- Gradient accumulation
- RLHF
- RLAIF
- Direct preference optimization (DPO)
Inference & decoding12 terms
- Token
- Tokenization
- Context window
- Key-value cache (KV cache)
- Prefill
- Decoding
- Quantization
- Speculative decoding
- Temperature
- Top-p sampling
- Beam search
- Test-time compute
Retrieval & representations12 terms
- Embedding
- Dense retrieval
- Sparse retrieval
- Hybrid retrieval
- BM25
- Approximate nearest neighbor search (ANN)
- Reranking
- Cross-encoder
- Bi-encoder
- Retrieval-augmented generation (RAG)
- Chunking
- Query expansion
Agents & integration12 terms
- Tool calling
- Orchestration
- Predefined workflow
- Agent
- Agent harness
- Model Context Protocol (MCP)
- Tool schema
- Structured output
- Sandbox
- Idempotency
- Human-in-the-loop
- Approval binding
Evaluation & measurement12 terms
- Precision
- Recall
- F1 score
- Calibration
- Perplexity
- Benchmark contamination
- Ablation study
- Distribution shift
- Out-of-distribution (OOD)
- Pass@k
- LLM-as-a-judge
- Goodhart's law
Security & privacy12 terms
- Prompt injection
- Indirect prompt injection
- Jailbreaking
- Least privilege
- Excessive agency
- Confused deputy
- Data poisoning
- Backdoor
- Model extraction
- Membership inference
- Differential privacy
- Pseudonymization
Serving & operations12 terms
- Time to first token (TTFT)
- Inter-token latency
- Throughput
- Continuous batching
- Prefix caching
- PagedAttention
- Tensor parallelism
- Pipeline parallelism
- Data parallelism
- Zero Redundancy Optimizer (ZeRO)
- Checkpoint
- Model drift
Learning paradigms & data12 terms
- Supervised learning
- Self-supervised learning
- Unsupervised learning
- Reinforcement learning
- Transfer learning
- Active learning
- Curriculum learning
- Synthetic data
- Weak supervision
- Multimodal learning
- Contrastive learning
- Federated learning
Strategy & governance12 terms
- Foundation model
- Frontier model
- Artificial general intelligence (AGI)
- Alignment
- Grounding
- Hallucination
- Explainability
- Interpretability
- Data provenance
- Model card
- Risk appetite
- Open weights