ResearchPaper5 min
SMILESGNN: Fusing SMILES and Graphs for Better Toxicity Prediction
Exploring how SMILESGNN leverages multimodal fusion to improve clinical toxicity prediction while offering interpretable insights into molecular structures.
Interpretability
Opening the box rather than describing the outside of it. The closest thing this field has to a mechanism-first discipline, and the source of most of the techniques that turn out to matter later.
Research1
Security1
SecurityDeep dive17 min
Abliteration: how one direction holds a model's refusals
Open-weight models ship with guardrails. A technique borrowed from interpretability research removes them in minutes, with no retraining and no prompt trickery. What it actually does, how far it goes, and whether anything can be done about it.
Sebastiaan with AI