It was a little meta for me
A 2026 book called “How to Talk to AI” came across my threshold today for 99p and I thought, “sure, why not?”
It’s good, actually. As I read, I discussed what he was saying with Grok. It was a very surreal afternoon.
I’m only on chapter two, but interesting stuff. People think LLMs are great at data but terrible at creativity. It’s the opposite – they’re bad on data but fantastic on creativity (if we define creativity as combining different ideas in novel ways). Lateral ‘thinking’ is programmed right in. (Grok agreed).
Give a model two completely different objects and ask how they’re related. It invariably comes up with some interesting and plausible answers that just wouldn’t occur to a human being.
He said that current models are all weighted heavily to the middle. Ask a model to pick a random number between 1 and 10 and it’ll pick 7. I asked Grok. It picked 7. Which means its solutions are corporate and lifeless, particularly when you hear them in large doses.
To get around this, you give the model a persona and ask it to answer as that person. It doesn’t matter who. The examples he gave were “A mad scientist with a penchant for flower arranging. A teenager who is allergic to phones. Elon Musk after smoking weed. A mute nineteenth-century witchdoctor.” It doesn’t change the accuracy of the answer. The act of giving it a role alone forces it to step outside its constraints.
Though they did get even better results on a design challenge when they asked it to answer as Steve Jobs.
Anyway, chapter 2. I’ll let you know how Grok and I get on.
Posted: June 16th, 2026 under personal.
Comments: none
Write a comment
Beware: more than one link in a comment is apt to earn you a trip to the spam filter, where you will remain -- cold, frightened and alone -- until I remember to clean the trap. But, hey, without Akismet, we'd be up to our asses in...well, ass porn, mostly.<< carry me back to ol' virginny