-
Post-Training Primer
A first-principles look at post-training for language models, tracing how policy gradients lead to PPO and GRPO. The post breaks down reward-to-go, advantage estimation, probability-ratio clipping, and group-relative rewards, while highlighting the practical differences between PPO’s learned critic and GRPO’s critic-free approach. It also covers implementation details and common pitfalls when putting these methods into practice.
-
Uncertainty in Deep Learning, p.2 (Methods)
In this post, we explores the main methods for estimating uncertainty in deep learning — from Bayesian inference and sampling-based ensembles to efficient deterministic and calibration techniques. We discuss how each approach captures different types of uncertainty, their computational trade-offs, and when to prefer one over another in practice.
-
Uncertainty in Deep Learning, p.1 (Intro)
Advances in deep learning have solved complex problems across multiple domains, aided by increased computational access. Yet, understanding the inner workings of these models remains a challenge, with uncertainty estimation providing some insights. In this blog post, we explore the basics of uncertainty estimation and how it makes AI smarter and more reliable.
- •
- 1
- •