Which of the following statements with regard to Large Language Models (LLMs) used in machine learning is/are correct ? 1. LLMs assign probabilities to the next possible words and then pick the one with the highest probability. 2. LLMs process data through mathematical optimization to minimise prediction errors. 3. LLMs produce unbiased outputs. Select the answer using the code given below :
- (a) 1 only
- (b) 1 and 2 only ✓ UPSC's answer
- (c) 2 and 3 only
- (d) 1, 2 and 3
Why the answer is (b)
• An LLM is a neural network trained on vast text to predict the next token: it computes a probability distribution over possible next words and, in basic decoding, selects the most probable one — statement 1 is correct.
• Training is an optimisation problem: the model's billions of parameters are adjusted by gradient descent to minimise the error (loss) between its predictions and the actual next words — statement 2 is correct.
• Statement 3 is wrong: because LLMs learn from human-written data, they inherit and can amplify its social, cultural and factual biases; bias mitigation is an active research concern, not a solved property.
• Sampling methods such as temperature and top-k mean the highest-probability word is not always chosen, but the probability-assignment principle in statement 1 holds.
• Hence 1 and 2 only, option (b).
Why the other options are wrong
- (a) 1 only
- Statement 2 is also correct: training minimises prediction error through optimisation.
- (c) 2 and 3 only
- Statement 3 is wrong: LLM outputs reflect biases in their training data.
- (d) 1, 2 and 3
- Statement 3 is wrong, so all three cannot be correct.
Asked in the GS Paper I of the UPSC Civil Services Preliminary Examination 2026, held on 24 May 2026. Question and answer key: Union Public Service Commission. Explanation: UPSC Answer Check.