Article

Pretty sad how base model style "pass in some tokens and we'll predict the next ones" APIs are so hard to get a hold of now, let alone actual base model checkpoints.

Using base models for turbocharged autocomplete is a really delightful experience, chat isn't always optimal UX!