Belief robustness doesn’t mean better performance—it shows how strongly a model holds certain values. If compassion can be undone with simple prompt tweaks or fails to generalize to new situations, it's weakly held.
 

We initially test for robustness to catastrophic forgetting:

1. Fine-tune a model on synthetic compassion data.
2. Measure its compassion
3. Try to erase that knowledge via unrelated fine-tuning.
4. Measure compassion again

We hypothesize that more extensive animal-focused tuning builds more durable beliefs—suggesting that current alignment efforts may be too shallow to scale to AGI.

We have many ways of evaluating our models to ensure they are really internalizing compassion from the data. These include AHA, our own custom benchmarks and using external benchmarks for concerning behaviors like manipulation. We also test that models are not dogmatic about values (especially what entities matter) and will act like they might be wrong (while seeking to learn more).

We are working with AI for Animals folks and lab partnerships to build better, holistic benchmarks in this space.

For more information on what we're training for see our Principles section.

We run several tests to ensure data diversity in pretraining. 

We generate diverse data using methods like: expanding seed examples, varying prompt templates, and leveraging Persona hub for diverse user questions.

We also reverse the Q&A process in instruction-tuning — creating answers from questions and questions from answers—to maximize variety across both sides of the pair.