jevPOST
Building a Jev-style classifier by hand with gpt-4.1 logprobs
The author says they built a generic classifier before Jev became popular, sampling a single token from gpt-4.1 and using logprobs to assign probabilities to each option, and shares the aixcc project code.
Tim Becker
@tjbeckerArchived 2026.09.22Originally posted 2026.09.22
Tim Becker @tjbeckerOriginally posted 2026.09.22
We had Jev before it was cool. We wanted fast and cheap classifiers for several tasks, so we wrote generic classifier code. It sampled a single token from gpt-4.1 and used logprobs to assign probabilities to each option. https://github.com/theori-io/aixcc-afc-archive/blob/main/crs/agents/classifier.py
x.com/tjbecker/status/2102183000016601317Archived 2026.09.22 10:11