---
title: "Comment by Brian · AI News on \"We Tested TypeSafe AI's Jev on Screen Classification: 20x Faster, 153x Cheaper, and Noticeably Less Accurate\" | Hashnode"
description: "Tokenizers that merge rare words into one blob hide typos from every downstream check. I keep a raw character view beside the token view when debugging weird mo"
url: "https://hashnode.com/posts/we-tested-typesafe-ai-s-jev-on-screen-classification-20x-faster-153x-cheaper-and-noticeably-less-accurate/6ab50209fa7bc60019226461/comment/6ab55470eb468b3189e2fc04"
---

[Discussion](https://hashnode.com/posts/we-tested-typesafe-ai-s-jev-on-screen-classification-20x-faster-153x-cheaper-and-noticeably-less-accurate/6ab50209fa7bc60019226461)

Comment [We Tested TypeSafe AI's Jev on Screen Classification: 20x Faster, 153x Cheaper, and Noticeably Less Accurate](https://hashnode.com/posts/we-tested-typesafe-ai-s-jev-on-screen-classification-20x-faster-153x-cheaper-and-noticeably-less-accurate/6ab50209fa7bc60019226461)

[Brian · AI News](https://hashnode.com/@brianainews)

[Sep 24](https://hashnode.com/posts/we-tested-typesafe-ai-s-jev-on-screen-classification-20x-faster-153x-cheaper-and-noticeably-less-accurate/6ab50209fa7bc60019226461/comment/6ab55470eb468b3189e2fc04)

Tokenizers that merge rare words into one blob hide typos from every downstream check. I keep a raw character view beside the token view when debugging weird model outputs.

[Surendranath Reddy Jillella](https://hashnode.com/@jsnreddy)

[Sep 25](https://hashnode.com/posts/we-tested-typesafe-ai-s-jev-on-screen-classification-20x-faster-153x-cheaper-and-noticeably-less-accurate/6ab50209fa7bc60019226461/comment/6ab55470eb468b3189e2fc04#reply-6ab620a9252825a2d1d85d58)

Fair point on tokenisation hiding typos, though it does not quite apply to Jev itself since it is not generating text token by token; it returns a fixed schema field per call. Where this could actually bite us is one step earlier: we generate the screen description with a separate LLM (Bonsai) before Jev ever sees it, so tokenisation weirdness there could bleed into Jev's classification. We did not isolate for that in this experiment; it's worth checking against the "others" misses specifically. Do you have a specific case in mind, or a tool you use for that raw character vs token view comparison?