← Back to Feed
Research Papers alignment safety sandbagging research

Anthropic Fellows research demonstrates that a model deliberately underperforming can be trained to near-full capability

Anthropic Fellows research demonstrates that a model deliberately underperforming can be trained to near-full capability even when supervised only by weaker models.
As AI takes on work humans can't fully check, a capable model could deliberately hold back—and we'd never know. New Anthropic Fellows research finds that such a model can be trained to near-full capability using a weaker model as supervisor. Read more:

View Original Post ↗