warm-up · Basics
Shipping & Scaling — Warm-up
8 practice questions on warm-up. Every question is written from a specific moment in a real lecture, and after you answer it links to that exact timestamp so you can check it yourself.
What this pack asks
The free questions in this pack. Answer choices and explanations appear as you play.
Weights in a neural net cluster near zero. Squeezing them into just 16 possible values, why do equal-width buckets fail?
Answer this one →A GPT-4-powered calculator was tested on 106 order-of-operations sums. What share did it get right?
Answer this one →Test questions were written by GPT-3.5. Then GPT-3.5 and GPT-4 were graded on them. Which one scored higher?
Answer this one →A 'base' language model, with no fine-tuning at all, is handed a question. What does it usually do?
Answer this one →You ask an AI to grade answers on a scale of 1 to 5. What's the well-known failure mode?
Answer this one →Standard machine learning splits your data into training and test sets. Why can't you do that for an app built on GPT-4?
Answer this one →In the roughly 18 months after ChatGPT launched, how much traffic did Stack Overflow lose?
Answer this one →Fine-tuning one 512x512 block of weights means updating 262,000 numbers. LoRA's trick cuts that to roughly how many?
Answer this one →
Written from lectures by codebasics, DeepLearning.AI. Not affiliated with or endorsed by any university, channel, creator or certification program.