Just a thought, perhaps worthy of discussion. Not sure how the costs would work out but if the dev could have JEV run alongside the regular System Two LLMs, imagine the impact on story-telling, RPGs etc? Anything that is longform text related.

EDIT: or an open source JEV-like System One model like Laya, though it can only read around a 1000 tokens rather than 32k - 64k tokens like the closed source, cloud-served TypeSafe JEV. But 1000 tokens feels plenty enough, considering the current model takes around 6k max.

[–] 1 point 1 month ago* (last edited 1 month ago)

If you see one of their recent posts, they address this. They didn't expect the AI coding helper to be so popular which is why they had to use a cheaper, dumber model to keep costs down.

Unavoidable, considering the nature of the site's monetisation, which is to say, none.

EDIT: and to be fair, the AI has become more economical. It no longer rambles and goes off on tangents and thinking loops.

What it is now, is more of a tool than a companion, which is honestly fine for a coding helper.

  • source