Paul Ford
Writer and co-founder of Aboard, who has published essays and stories on Ftrain.com since 1997 and writes about software, AI and vibe coding.
“A huge part of working with AI is trying to figure out how to use less of it to get more out. Like, what is the least of this I can use to get really good outputs?”
The Aboard Podcast · AI Builders · September 2026
Rich has compared large language models to using a riding mower on a tiny lawn. Paul agrees and adds that the good models are getting expensive, which sets up their discussion of a much smaller, faster model that picks from fixed options instead of chatting.
Transcript
You could use a push mower just for that. And I think
that's where we are today. A lot of things we ask of AI.
We keep, and actually a huge part of working with AI is trying to figure out how to use less of it to get more out. Like, what is the least of this I can use to get really good outputs? It's getting expensive. It is expensive, like to use the good stuff. So, okay. Okay. Now, and this is, so a huge story shows up in the AI world. A huge nerd story. If you're not in the AI world, you probably haven't heard of it. But John is a reporter working with this. He's like, what the hell is this thing? And it does. Is it huge? Yeah. Yeah, it's a really big story. So Jev, J-E-V, which is sort of for Javon's Paradox, which is that famous paradox that, you know, they thought they would use less coal, but they used more coal.
Yeah.
Okay. Shows up, and it's from this company, TypeSafe. And it is, they call it a system one model. But here's what's different: you don't talk to it, you give it a set of options. You say, and the options are kind of yes, no, multiple choice, probabilities, things like that, numbers. And you say, okay, Jev, and you prompt it just like you would a LLM. You give it, you're like, hey, Jev, I got this document for you.
Yes.
More from The Aboard Podcast
