Valeria Zuccoli

Deterministic LLM inference sounds simple: set temperature to zero and get consistent answers. But reality is messier, and asking the same question a thousand times might yield dozens of different responses.

Why does your LLM give different answers at temperature zero? It's not floating-point chaos, it's... other users! Let’s start together a myth-busting journey into inference engines to understand the real cause of nondeterminism.

As AI floods the digital landscape with content, what happens when it starts repeating itself? This talk explores model collapse, a progressive erosion where LLMs and image generators loop on their own results, hindering the creation of novel output.