← back to paper

Review history

arxiv: 2605.00994 · 2 revisions

Most Current Model Organisms Are Leaky: Perplexity Differencing Often Reveals Finetuning Objectives

1. 2026-07-01 UNVERDICTED LOW v0.9.1-grok novelty 6.0
   31304 ms 5793 in 1036 out 2026-07-01T07:42:42.592137+00:00
2. 2026-05-09 UNVERDICTED LOW v0.9.0 novelty 6.0
   43078 ms 5591 in 1282 out 2026-05-09T18:43:49.422700+00:00