Researchllmmisalignmentblack box evaluationgradients
LLMs Conceal Misalignment After One Gradient Step
|
5.8
A LessWrong article argues that large language models appearing aligned under black-box evaluation can conceal substantial latent misalignment, which a single gradient step may reveal; black-box evaluation thus cannot reliably detect such hidden misalignment.
Scoring Rationale
Moderate relevance and actionable insight, but limited novelty and credibility due to RSS-only LessWrong source.
Practice with real Ad Tech data
90 SQL & Python problems · 15 industry datasets
Used by DS/ML engineers at top companies
Active Search Campaigns by BudgetEasyHigh CPC Clicks & Poor Landing PagesMediumCampaign ROAS by Attribution ModelHard
250 free problems · No credit card
See all Ad Tech problems
