machine learning

Inside LLMs: How Pre‑Training Shapes What ChatGPT Knows

by

The foundation of any Large Language Model (LLM) lies in a process called pre-training. This is where the model learns how language works by processing an immense volume of human-generated text. Pre-training is self-supervised, non-interactive, and results in a static model: it defines what the model “knows”, and more importantly, what it doesn’t. Pre-training teaches … Read more

Inside LLMs: RLHF, RLAIF & the Evolution of Model Alignment

by

While pre-training equips Large Language Models (LLMs) with a broad statistical understanding of language, it does not make them helpful, safe, or aligned with user expectations. Left in their raw form, these models can be verbose, biased, evasive, or simply unhelpful, even when technically accurate. To bridge the gap between linguistic fluency and user alignment, … Read more