What AI chatbots are doing under-the-hood, LLMs from scratch part 1

Link: https://www.gilesthomas.com/2025/08/what-ai-chatbots-are-doing-under-the-hood

Context

This is a great series, I am reading more about it this weekend. LLMs are things that produce logits, which is like a list of probabilities from a vocabulary and for each token the highest probability token should have been chosen, however it is random the temperature determines how random the selection is, 0 means choose the first(highest probable, no creativeness), 1 means choose any random one. Between these 0 and 1 you can experiment to find the sweet spot for your needs.

Source: techstructive-weekly-60