Researchers Showcase Six Papers at ICML 2026 in Seoul

Our ICML paper on random factors in emergence reveals that whenever we observe ABRUPT "emergence" of a capability at scale, we're actually sampling from a multimodal distribution which is GRADUALLY changing the probability of the capability with scale.

Our ICML paper on verbalization techniques for interp reveals that current verbalizers (and verbalizer evaluations) fail to consider if an interpretation reflects the verbalizer's background knowledge rather than the subject model's processing.

Our ICML mechinterp workshop paper demonstrates how feature geometry can lead to model failures, and analyzing that geometry can help us to efficiently build adversarial test sets based on concept combinations.

Our ICML oral (Tuesday!) position paper argues that empirical science of DL and future model improvements both rely on understanding the training process, not just analyzing or manipulating a fully trained model.

Our ICML position paper argues that interpretability not only SHOULD be actionable, but CAN be---just follow our framework for targeting actionable insights.

Our ICML HiLD workshop paper shows that the reason why bigger models learn more complex tasks is because they are able to saturate the gradients for easier tasks; different tasks are competing for the same parameters and gradient mass.