The evidence behind the guides

Sources

Find a study, check what it supports, and follow it back to a concern.

TopicCollection

Sources for Political ideology in RLHF preferences

2 sources

Guide citations are still under editorial review.

Baumgärtner and colleagues Research preprint

Best-of-Venom: Attacking RLHF by Injecting Poisoned Preference Data (original source, new tab)

Experiments manipulate preference datasets and model sentiment. Scope is the experimental setup, not an estimate of commercial prevalence.

Santurkar and colleagues · ICML 2023 Research study

Whose Opinions Do Language Models Reflect? (original source, new tab)

Compares model responses with US public-opinion groups. It does not isolate training data as the sole cause of differences.

Sources retain their original publishers’ rights. Inclusion does not imply endorsement. Saved sources stay in this browser on this device. Dates and source descriptions come from the existing editorial records.