10 Comments
User's avatar
eileen's avatar

You can actually adjust the model parameters to produce more creative and variable results.

In your prompt you tell the model to “set the temperature” and set it to value between 6 and 2 (max.)

You will have a higher chance of hallucination, but also more unusual model responses.

Expand full comment
Paul R. Pival's avatar

Say, for your Jan example (https://docs.google.com/document/d/1raFl1tsS4bSW0yjbjtBZg5u7wHN9Ep_IOaIkOkBhafE/edit?tab=t.8ig70slmrtvq) Is there a model you prefer? Do you recall which one was in play for this response?

Expand full comment
Jeremy Caplan's avatar

Hi Paul, Jan-v1 (4B-Q4_K_M) is what I'm testing and what I used for this example. Works well for me and I can recommend it, though I haven't done extensive comparisons with all the other model options. Curious if you or others have different preferences based on what you've tried.

Expand full comment
Paul R. Pival's avatar

Thanks! I haven't extensively used local LLMs, but it's a goal for this coming semester, and I have a 19-hour flight this weekend, so perfect timing! Currently limited by hardware - the best machine I have available is a M1 MacBook Air with 8GB of RAM, and it doesn't seem to like the model you just mentioned. :-/

Expand full comment
Rebecca Caroe's avatar

Cracking. Love the examples

Expand full comment
janet grau's avatar

Very helpful. Thanks!

Expand full comment
The 2020 Report's avatar

Awesome article

Expand full comment
Jon (Animated)'s avatar

Loved this. Your great guide makes AI feel playful and liberating rather than mechanical.

Expand full comment
Haikufarmgirl's avatar

I love these tips for tinkering with prompts!

Expand full comment
Raj Sidhu's avatar

Hey Jeremy really cool tips jere. Gonna try these out tomorrow and see what i van create. 💥🌟

Expand full comment