Member-only story
Girls just want to train and scale large language models
Content Design leader at Gemini on how to build eloquent LLMs.
It’s been almost a year since the last Button conference, and I’m still not finished reviewing my notes.
For those who don’t know, Button is basically the Met Gala of content design. Instead of outfits, UX celebrities compete with Google Slides presentations.
Today, I’m finally sharing some notes from Morgan Marie Quinn’s talk, a content design (CD) leader at Gemini. I’ll also include some of my own thoughts along the way. By the way, her talk was called “What the f#@& does Content Design have to do with LLMs?” — which signals that Quinn is not only a queen by her last name but also by nature.
At the time, Google’s LLM — Large Language Model — was still known as Bard, so, technically, no one can blame me for putting a gorgeous Frans Hals painting on the story’s cover.
LLM is not born, but made
One famous book about queens, kings, and pirates (you don’t know the name of it yet because I haven’t finished writing it) has a passage that says:
“A king is seldom born a good ruler. It is training that makes him so.”
The same is 100% true for LLMs. Quinn compares training Gemini/Bard to teaching a baby to talk. As with a small kid, it involves:
- a lot of reading,
- exposing them to different social settings,
- interpreting and reflecting the words back,
- talking to them (all the time).
First things first — it’s not the same as designing a chatbot
One of the mistakes CDs make is thinking that designing an LLM is similar to designing a chatbot. But! The outcome of a response from a chatbot is always the same. On the contrary, LLM is dynamic, and you cannot predict its response to the T.
Sometimes, the responses of a young LLM don’t make any sense. These slips are…
