Overview of AI
1. What is the relationship between the amount of training data and AI model capabilities?
Correct! Different AI models require vastly different amounts of training data - simple models might need hundreds of images, whilst ChatGPT-4 was trained on 45 terabytes of data. More data typically enables more sophisticated capabilities.
Not quite. The course shows that more complex models like ChatGPT-4 need vastly more training data (45 terabytes) compared to simpler models (hundreds of images), and this scaling enables more sophisticated capabilities.
2. How do large language models like ChatGPT fundamentally work?
Correct! LLMs work by breaking text into tokens, finding unique positions for words in high-dimensional space based on context, and predicting the next word using mathematical models called transformers. GPT stands for "Generative Pretrained Transformers".
Not quite. LLMs use mathematical models to place words in high-dimensional space and predict the next word based on patterns learned from training data, rather than storing or retrieving exact text.
3. What is the relationship between AI, machine learning, and deep learning?
Correct! AI encompasses the whole field of getting computers to mimic human behaviour, machine learning is a subset that learns from data rather than hard-coded rules, and deep learning is a subset of ML using deep neural networks inspired by the brain.
Not quite. AI is the broadest field, machine learning is a subset of AI focusing on learning from data, and deep learning is a subset of ML that specifically uses neural networks.
4. What significant breakthrough did Google DeepMind's AlphaFold achieve in biological research?
Correct! AlphaFold revolutionised biology by accurately predicting the 3D structure of proteins from their amino acid sequences, solving a problem that had challenged scientists for 50 years and opening new possibilities for drug discovery and disease understanding.
Not quite. AlphaFold's breakthrough was in protein structure prediction - determining how proteins fold into their 3D shapes based on their amino acid sequences, which had been an unsolved problem for decades.
5. Why might an AI image recognition model incorrectly classify a slightly noisy image of a cat?
Correct! AI models work with the actual RGB numerical values of pixels. When noise changes these values, even if humans can still see it's a cat, the mathematical input to the model changes, potentially leading to misclassification.
Not quite. The issue is that AI models work with numerical RGB values, so when noise changes these numbers, it affects the mathematical input the model receives, even though humans can still recognise the image.
Complete the quiz to see your results.