“13B didn't work really well. So now we're kind of on…” — Software Engineering Daily at 5:30
“13B didn't work really well. So now we're kind of on average at like the 70B scale. And that's where we see it's easy to deploy, it's cheap. And since we're so specialized in that one use case, we can get rid of a lot of the other generic use cases built into the model, right? And kind of read and train a lot of it…”
0:00
42:20