What it does
On 400 recognizable sketches, he reports about 35% accuracy and a strong airplane bias. This is a test outside Jev’s intended text tasks, not evidence that it can see images.
Benchmarks & research
TypeSafe’s Jev Can’t See. I Made It Guess What I Drew Anyway
Bartosz Mikulski turns drawings into SVG coordinates and asks text-only Jev to choose which object each one depicts.
Bookmark: TypeSafe’s Jev Can’t See. I Made It Guess What I Drew Anyway
Keep this in your collection.
Leave a noteWhat would you try with this?
: TypeSafe’s Jev Can’t See. I Made It Guess What I Drew Anyway
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
How you can use it
For your own AI comparison, borrow the study's testing rules, not its sketch recognizer. Pick a fixed sample and decide what counts as close enough before seeing results. Give both models the same examples, then examine cases where one is right and the other wrong.
Start with the article's test and prewritten-rules sections. Its sample included only drawings already recognized by Google's system, so the results do not cover every sketch. The author says Jev is not useful for practical sketch recognition; this is a comparison method, not a drawing tool to install.