✨ If you're new to prompting, we recommend checking out our Beginner's Guide to Prompting. It's a great way to get started and learn the ropes.
Understanding Weighted Prompts in AI
When it comes to artificial intelligence, prompts are important in helping models produce the desired results. As AI systems get better, users can interact with these models in more advanced ways. One way this is happening is through the use of importance scales and weighted prompts. But what are these scales exactly, and how do they change the way AI models respond to inputs?
Understanding the Baseline
To make things easier to understand, let's talk about a starting point. When you create a prompt, each word has the same weight of 1.0. This means that the AI model sees all the words as equally important. But, if you want to highlight some words as more important or less important, you can adjust their weight using the importance scale. This will change how the AI model perceives each word compared to the starting point.
Unraveling the Importance Scale
The importance scale is a system that lets users rank the significance of words or phrases in a prompt. By using the scale, users can indicate which parts of their input should receive more attention than others.
Think of it like a scale where words or phrases are assigned values based on how important they are. Words with higher values are considered more significant and receive more attention from the model, while those with lower values receive less attention accordingly.
How to Assign Importance
To make Kaiber and other AI models interact better with certain parts of a prompt, users can use weights. Weights are multipliers that show how much importance a model should give to a specific word or phrase. Here is a quick breakdown on how to use these weights:
A (word): If you put a word in one set of parentheses, the model will pay more attention to it, about 1.1 times more. This makes the word a bit more important in the model's calculations, but not by too much.
a ((word)): Double parentheses increase the emphasis on the word. The attention to the word is increased by a factor of 1.21, which is essentially 1.1 multiplied by itself.
a (word:1.5): Specifying a factor directly gives better control. It makes the attention to the word "word" 1.5 times stronger.
a (word:0.25): Using a factor below 1 lets users decrease how much the model pays attention to a word. In this case, the attention is decreased by a factor of 4, which is the reciprocal of 0.25.
The Implications of Using an Importance Scale
This scale lets users create prompts with a detailed order of importance. For example, in storytelling, you may want to make sure that the theme of "redemption" is more important than a small detail like "rainy weather". Using the importance scale, you can give this specific instruction to the AI model.
Example Prompt: young girl, on top of tall skyscraper, (redemption), tears on her face, clenched fist, (rainy weather: 0.25)
Weighting Strategies:
To create a clear order of importance in the prompt, assign weights. When multiple words or phrases have the same weight, the AI will treat each of them as equally important, so follow the strategy below.
Begin by identifying the most important elements, give them the highest weights
From there, systematically reduce the weight as you progress to less significant elements
If you notice something odd or uncommon in your results, it might suggest that a certain word or phrase has been given too much importance.
💡 To get more consistent and reliable results, it's better to stay under a threshold of 1.8.