Which of the following statements with regard to Large Language Models (LLMs) used in machine learning is/are correct ?
- LLMs assign probabilities to the next possible words and then pick the one with the highest probability.
- LLMs process data through mathematical optimization to minimise prediction errors.
- LLMs produce unbiased outputs.
Select the answer using the code given below :
Detailed Explanation:
Statement 1 — Correct. LLMs analyze input context and calculate a statistical probability distribution for the next possible word (token). In their basic form (greedy decoding), they pick the word with the highest probability. While real-world applications use controlled randomness (top-k sampling, temperature scaling) for creativity, the foundational mechanism is probability-based selection.
Statement 2 — Correct. LLM training relies on mathematical optimization:
- Algorithms like Stochastic Gradient Descent (SGD) and backpropagation iteratively adjust billions of model parameters
- The explicit goal is to minimize a "loss function" — a mathematical measure of prediction errors vs. actual training data
Statement 3 — Incorrect. LLMs do NOT produce unbiased outputs. Since they learn from massive human-generated datasets (internet text), they inherit, reflect, and sometimes amplify historical and societal biases — including gender, racial, and cultural stereotypes — present in training data.
Key Trick: Statement 3 uses an absolute term — "unbiased" — which is a classic UPSC red flag. No AI model trained on human data can be completely unbiased.
Question 1 of 2 Machine Learning
Practice PYQ questions from this topic across all years
First question in this topic
Consider the following statements: I. It is expected that Majorana 1 chip will enabl...