usmananwar.bsky.social @usmananwar.bsky.social · 11/11/2024Transformers are REALLY good at in-context learning (ICL); but do they learn ‘adversarially robust’ ICL algorithms? We study this and much more in our new paper! 🧵 121
