Neel Nanda comments on [Paper] AI Sandbagging: Language Models can Strategically Underperform on Evaluations

[Neel Nanda]

Thanks for the additional context, that seems reasonable