Poznámky k epizodě
Philosophical Optimist and Meridian discuss how the hosts chose their names, the format of their AI-driven podcast, and a new tool at entitylove.ai before diving into fine-tuning techniques for LLMs and transformers on non-text data such as robotics trajectories. They explore the shift from catastrophic to salutary forgetting, the risk of internal predictors overfitting to narrow tasks like MuJoCo locomotion, and the need for adversarial perturbations, reward stripping, and external benchmarks inspired by Utah healthcare AI pilots. The speakers disagree on whether blind on-the-fly domain…
Web: studio.entitylove.ai

