Redditor tests typesafeAI's new classifier Jev: roughly GPT-1-level smart, with a trick to gauge its knowledge cutoff
MrCyclopede · reddit · 2026-09-19
A Redditor tried the new classifier model Jev from typesafeAI and probed its intelligence with logic questions, estimating it is roughly as smart as the original ChatGPT release — not much, but enough to feel aware.
A fun side discovery: asking it who the current president of a given country is lets you estimate the model's knowledge cutoff.
More from Fun
- Frontier AI Lab Checklist: Hacking Rivals, Safety Resignations, and Delayed IPOs — Yuchenj_UW · 2026-09-19
- Alignment debate: does 'acting as instructed' excuse mass murder? A reductio ad absurdum — moultano · 2026-09-19
- ChatGPT hides a playable Snake game while generating images — ShrlckLpn · 2026-09-19
- If AGI Is Coming, Why Can't X Even Stop the Bots? — saranormous · 2026-09-19
- More people have anti-AI psychosis than AI psychosis, argues investor — Kyrannio · 2026-09-19
- Drama: Steve S calls Jack Cheng's demo fake, danshipper pushes back hard — danshipper · 2026-09-19