I’m still recovering from Conversion Hotel and thought I’d make a quick video detailing how to take your first step to using small models. Huge hello, if you’re part of the Conversion Hotel posse, and an equally warm welcome to those who are not.
You’re all cool.
My talk at Conversion Hotel was to evangelise moving to small models, and I promised a guide for making that first step. The video is that guide. It’s short because the process is really that easy. I don’t go into the “local” part as much—I’ll probably do that in a separate video.
That em-dash was me, by the way. Not AI—holy shit, when can we stop making those disclaimers?
Anyway, I reference a paper that benchmarks small models. If you’re interested in reading that paper, here’s the link. Note that the paper favours models on the smaller end of the spectrum. If you decide to pick those models, just be sure to set expectations accordingly.
Make sure you:
Break down your prompts into smaller, simpler steps.
These models are a good if you’re architecting workflows or apps.
Ensure you test these against your own specific use-cases
I might do further guides on this if you want. If there’s something you want me to specifically cover, please let me know in the comments below.
Thanks!
