← Toutes les offres
D

Alignment 'train/test splits' should be sacred

draganover

Global
Publié le 16 novembre 2026

Description du poste

Tl;dr: Alignment techniques can be used for training alignment or for testing it, but they cannot be used for both! As we build out the set of alignment techniques available to us, we should enforce that some are judiciously reserved only for evaluating alignment. See also: Validating against a misalignment detector is very different to training against one. Here is a guiding principle regarding alignment-minded AI training: whatever you do, you must maximally avoid teaching the model to take misaligned actions which are invisible to the user. In fact, you should be shamed if you use training techniques which encourage covert misalignment. Here is a separate, guiding principle of machine learning: whatever you do, you must maximally avoid training the model on the entire dataset rather than holding out a test split. In fact, you should be shamed if you report results which are obtained by training on the entire dataset rather than on the train split. I want to argue that these are the same principle. Training against all of your alignment assessment techniques leads to the same failure as training on your validation data in traditional machine learning: you potentially get better performance, but you sacrifice all ability to verify your model’s behavior. Consider two types of alignment techniques: • Some techniques sculpt the model’s alignment. This includes techniques such as:• Steering and other inference-time interventions• SFT generalization techniques (midtraining, pretraining filters)• RL generalization techniques (RLHF, RLAIF, beneficial RL, GRAFT, inoculation prompting)• Prompt engineering• Inference time safety classifiers• Some techniques assess the model’s alignment (aka, monitoring). This includes techniques such as:• Human analysis of model outputs• CoT monitoring• Activation monitoring• Broader interp tools like SAEs, transcoders, etc.• Evals for whether models can control their activations• Pre-fill analysis• Honeypots• Universal jailbreaks (to study