Jev
TypeSafe's System One model, v1.14
What we asked Jev
NotHotdogBench v1
Not Hotdog
We set out to prove that Jev, TypeSafe's System One model, can't tell a hotdog from a dachshund. Then we ran it.
Inspired by Jian-Yang's SeeFood app from Silicon Valley.
Every Jev and MobileNet answer on this page is a real result, recorded on 15 Oct 2026. Paul's picks come from a random number generator, because Paul died in 2010.
Speed round
14 photos. 3 judges. No retakes.
Jev is SOTA*
Dots per photo, left to right: Jev, MobileNet, Paul
Right
Wrong
Debatable, not scored
Dashed red box: where Jev says the hotdog is
Jev is SOTA on NotHotdogBench*
Accuracy on the photos with an agreed answer, %
Full results
How each judge counts as saying HOTDOG is listed under the table.
| Judge | Accuracy | Median time | Runs on |
|---|
Jev: says HOTDOG when it answers yes with at least 50% probability.
MobileNet: says HOTDOG when its top guess is the ImageNet class "hotdog".
Jev's time is a full API round trip from a laptop. MobileNet's is model time only, on the same laptop's CPU.
How each judge worksJev's questions, MobileNet's guesses, Paul's boxes
Same photo, three judges, slow motion.
MobileNet
Free image classifier from 2018
Top-5 ImageNet guesses
Paul
The famous octopus from the 2010 World Cup
How Paul judges
Two boxes go into the tank, one labelled HOTDOG and one NOT HOTDOG. Paul eats from one. Since 2010 a random number generator (seed 2010) eats on his behalf.
Every answerAll 16 photos, every judge's call
| Photo | Answer | Jev: hotdog? | Jev: score 0–2 | MobileNet's top guess | Paul |
|---|