Feed

Overfitting happens when an AI memorizes its training data instead of learning patterns — perfect on familiar examples, but failing on new ones.

Training data is the collection of examples an AI learns from — its abilities, knowledge, and blind spots all come directly from what it was shown.

In AI, inference is when a trained model makes predictions from new input — every chatbot reply and generated image is the result of inference.

LoRA (Low-Rank Adaptation) customizes a large AI model by training a tiny add-on — under 1% of its size — instead of retraining the whole thing.

Quantization shrinks an AI model by storing its numbers with less precision — letting powerful models run on ordinary laptops and phones.

A context window is the amount of text an AI can read and remember at once — its short-term memory. Learn how it works and why size matters.

A vector database stores data by meaning using number lists called vectors — the technology behind AI search, recommendations, and chatbot knowledge.

Chain-of-thought prompting asks AI to show its step-by-step reasoning before answering — a simple trick that makes answers to complex questions far more accurate.

Few-shot learning: show AI a few examples and it follows the pattern. Learn how to write example prompts that get accurate, on-style results.

Zero-shot learning means AI can do new tasks with zero examples. Learn how instruction-following works, when it shines, and when to add examples.