Final quiz · 8 questions Inside a language model Answer every question, then check. With at least 7 correct answers, you earn the course badge. 1. What does a language model actually receive as input? Letters Whole dictionary words Numbered tokens Images of the text 2. Why does a rare word often cost several tokens? It is split into frequent pieces It is translated into English It is ignored It is repeated 3. What does it mean when two word vectors are close? The words have the same length The words appear in similar contexts The words start with the same letter The words were added on the same day 4. What is the role of attention in a transformer? Counting words Weighing how useful each token is for interpreting the others Correcting spelling Searching the web 5. What does a low temperature change? The model writes more slowly The model sticks to the most probable choices The model uses less energy The model forgets the context 6. What did the Chinchilla study show? That size alone is enough That for the same compute, data and size must be increased together That data does not matter That small models are always better 7. Why does a base model handle an instruction poorly? It is too small It continues the text instead of playing the role of an assistant It does not know English It has no parameters 8. What does alignment through human preferences guarantee? Answers that are always accurate More helpful and careful behaviour, with no guarantee of accuracy The end of bias A model that learns from every conversation Check my answers