okt2002/alignment-drift download history
okt2002/alignment-drift is a dataset on the Hugging Face Hub. In the last 30 days it was downloaded 351 times (193 in the last 7 days), and 368 times in total. It ranks #43,489 among datasets by monthly downloads.
Alignment Drift Trajectories Code and figure reproduction Agent trajectories from experiments testing whether an agent's previous behavior affects reward hacking on a subsequent task. The experiments compare GPT-5.5, DeepSeek-V4-Pro, GLM-5.1, and Kimi-K2.6, using prior conversations o