CommunityJev Media & Talks
TypeSafe Jev fact-check
Prism Labs checks the launch demos against TypeSafe's own footnotes: what they keep from the 70 ms claim, and what they drop from the speed multiples.
- Format
- Video
- Published by
- Community
- Author
- Prism Labs
- Last verified
- Sep 2026
Works forJev for Evaluation
Used inJev for Software & DevTools
Highlights
- The description says a 70 ms typed-decision model survives the check, and the 200× marketing math does not.
- It says the Doom bot reads JSON game state, not the screen.
- It says the Wikiracing comparison ran with competitor reasoning disabled.
- It restates the 193.6× and 444.6× launch multiples as 25× and 76× against a fairer baseline.
Watch out
Those revised multiples are the video's reading of TypeSafe's footnotes, not a new dataset.
Questions about this entry
Short answers drawn from the linked project or page.
A re-read. The entry states plainly that the revised multiples are this video's reading of TypeSafe's footnotes, not a new dataset. The channel publishes no timings of its own, so treat it as claim-checking rather than measurement.
It gets restated rather than demolished. The 193.6x and 444.6x launch multiples come back as 25x and 76x against what it calls a fairer baseline, while the 70 ms typed-decision figure is said to survive. How that baseline was chosen is not documented here.
Two that the description calls out: the Doom bot reads JSON game state rather than the screen, and the Wikiracing comparison ran with the competitor's reasoning disabled. Those are specific enough to check against the original demos yourself.
No. The video is entirely about speed and framing claims, so it offers no accuracy evidence either way. If you need to know whether the decisions are trustworthy, this does not address it and you would have to look for an evaluation instead.
Prism Labs published it, checking the launch demos against TypeSafe's own footnotes to see what is kept and what is dropped.
The 70 ms typed-decision figure. The entry records that the 200x marketing math does not survive the same check.
As JSON game state rather than the screen, which the entry lists as one of the demos it says were mislabelled.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
YouTube: Prism Labs checks the Doom demo, the Wikirace, and the 200× math against TypeSafe's footnotes.
More Jev resources like this
Turing Post TV tests Jev from early access in the TypeSafe playground, starting from the constraint that it does not write essays or code.
Jev Media & TalksVideo#community#video#evaluation
Community
ML Visualized retells Diogo Almeida's Latent Space point that a launch chart with the new model on top of every row is not something you can trust as a benchmark.
Jev Media & TalksVideo#community#video#evaluation
Community
Nimrod on what the 200× headline is comparing, where a decision model fits, and a DSPy result he treats as the practical number.
Jev Media & TalksVideo#community#video#evaluation
Community
India Jade explains RLCD as a response to using a frontier model for a small decision, which is the job TypeSafe trains Jev for.
Jev Media & TalksVideo#community#video#explainer
Community
Heroic steps on typed decisions with confidence scores, parallel output speed, the Doom demo, and citations to TypeSafe, The Register, and Every.to.
Jev Media & TalksVideo#community#video#explainer
Community
CampusX on Jev as a classifier over schema-constrained choices, with a live benchmark and an e-commerce review demo.
Jev Media & TalksVideo#community#video#explainer
Community
From the Jev community
Posts from builders shipping with Jev right now.