AI Glossary

Plain-English definitions for the benchmarks, techniques, and jargon that decide which AI model wins. Built to demystify the daily ranking.

A

B

Benchmark
A standardized test for comparing model capabilities.

C

D

Distillation
Training a small model to mimic a big one.

E

Embeddings
Numeric vectors representing text meaning.

F

G

GPQA
PhD-level science questions resistant to web search.

H

I

Inference
Running a trained model to generate output.

J

Jailbreak
Prompt that bypasses safety guardrails.

K

Knowledge Cutoff
Date after which the model has no training data.

L

M

O

Open Weights
Model parameters are publicly downloadable.

P

Prompt Engineering
Crafting inputs to get reliable outputs.

Q

Quantization
Storing weights in fewer bits to shrink and speed up models.

R

S

T

V

Vision-Language Model (VLM)
Multimodal model specialized for image + text.