Empirical Science and Applied Research on Generative Models

Experiments on the open questions in generative models.

The most important questions about generative models are still open — what post-training actually changes inside the model, when alignment holds under pressure, whether RL improves reasoning or just its appearance. Applied Models runs focused experiments to find out.

The model landscape is moving faster than our understanding of it. That gap is the work. Seven directions, each with concrete questions worth pursuing precisely — and enough control over the experimental setup to actually answer them.

Models and datasets: Hugging Face organization