Skip to content
anythingai.space
Generative AIbeginner12 min · 9 steps

Prompting, Fine-Tuning and RLHF

Four levers for getting what you want from a pre-trained model — zero-shot instructions, few-shot examples and prompt patterns, fine-tuning on your data, and the three-stage RLHF pipeline.

THREE RUNS, THREE SHAPES OF ANSWER“It's a mixed review overall, because…”“7/10”“Good for whom?”UserPrompt“Is this review good?”Pre-trained LLMfrozen weightsOutput?
requesterrortokenresponsesignaldataheavy work
1 / 9
ProblemStep 1 of 9

Same model, vague prompt, unpredictable output

A pre-trained model can do a lot — but it only sees the text you give it. Ask “Is this review good?” and you might get a paragraph, a score, or a question back.

The model's weights aren't the problem. The input is under-specified, so the model guesses at the task, the format and the tone.

Watch outDefault, unstructured prompts are simple to write and the least effective way to use a model.

Lesson 3 of 8 in Generative AI