🔥 ChatGPT Tips Tuesday #10: "Weigh" your prompt

💡 ChatGPT Tips for the day

Thanks to all the people who voted in the last issue. By popular demand, I am going to talk about adding some weights to your prompt from ChatGPT.

Just a disclaimer, this is not something I come up with myself. Here is where I have heard this concept:

For this issue, I plan to test if this approach works and explore other potential use cases where we can use this.

Following the suggestion, I tried the following prompt to ChatGPT

If I prepend [value] before a line this is its importance [0.5] being half as important as [1.0] and [10] being [10] times as important. If i do not, it is [1.0] importance.

Do not treat them separately. Use them all to influence the entire response. Do not output weights.

Describe juggling in 20 words.

[0.2] use words that start with R
[10] use words that rhymeTry this in your next prompt

Here is what has come out from GPT-3.5. I mean it is a good explanation but probably lacks a bit of rhyme.

However, in GPT-4, the result looks much better. It flows much better.

Also, I have to try the Snoop Dogg one. Pretty smooth I must say 😎 

In conclusion, here is what I found:

• Weights work!

• It works better with GPT-4: GPT-3.5 do seem to ignore the weights assigned. Not sure if somehow shortening the prompt text would make it work on GPT-3.5

Potential use cases I can think about now:

• A Midjourney prompt generator where you can generate outputs based on different tones and style settings, e.g. tuning the expression of the people, and the style of the output.

• An email generator where you can generate emails based on different tones, e.g. being formal or informal, being persuasive or not

Try this in your next prompt

If I prepend [value] before a line this is its importance [0.5] being half as important as [1.0] and [10] being [10] times as important. 

If i do not, it is [1.0] importance.Do not treat them separately. Use them all to influence the entire response. Do not output weights.

{your prompt, e.g. image generation}

[level of weight] {1st control/limitation of the prompt, e.g. style setting}
[level of weight] {2nd control/limitation of the prompt, e.g. style setting}
...
[level of weight] {nth control/limitation of the prompt, e.g. style setting}

🔤 Word of the day

Unsupervised learning

n. is like a self-taught artist who learns to paint without any formal training. Instead of being given specific instructions, the artist explores and identifies patterns on their own.

In the world of artificial intelligence, unsupervised learning algorithms are similar. They explore raw data—like a massive photo album—and find patterns or groupings without being told what to look for. For example, they might sort the photos into groups like 'beach photos', 'family gatherings', and 'selfies', all on their own. This helps us understand and make sense of large amounts of data.

To explore further about this concept, check this page on Wikipedia

🖼️ Image of the day

Buddy, I feel you 😂 

💕 Share the love and get some cool avatars!

I am doing weekly draw to get people referring the newsletter to others with some cool avatars as a way to say thank you.

If you can refer at least 3 of your friends, you would be in this week’s draw of 3 to get 3 cool avatars using 3 of your chosen styles.

So if you enjoy this newsletter and want to get some cool avatars for yourself, here is your link to share your love of the Prompt Craftsman with your friends.