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:
Answer & explanation
Answer: (b) 1 and 2 only
An LLM is a statistical text predictor: it scores possible next words by likelihood and produces the most probable continuation, and it is built by training that tunes its parameters to reduce prediction error. Because it learns from human-written text, it inherits that text's biases, so its outputs are not unbiased.
- ✓ 1. An LLM works out how likely each possible next word (token) is and generates the most probable continuation; NIST describes LLMs as predicting the next token or word in a sentence or phrase.
- ✓ 2. Training adjusts the model's parameters so that its errors against an objective on the training text become as small as possible; NIST defines the training stage as one in which a model learns parameters that minimize its error against an objective function.
- ✗ 3. LLMs learn from human-written data, and NIST warns that bias becomes ingrained in automated systems, including generative AI. A UNESCO study found LLMs producing gender bias, homophobia and racial stereotyping.
Remember · LLMs predict the most probable next words and are trained by optimisation to cut prediction error. They inherit bias from training data, so outputs are never automatically unbiased.
Sources
- NIST AI 600-1, Artificial Intelligence Risk Management Framework: Generative AI Profile (2024): how LLMs generate text ↗ “they generate outputs that approximate the statistical distribution of their training data; for example, LLMs predict the next token or word in a sentence or phrase. … Bias exists in many forms and can become ingrained in automated systems. AI systems, including GAI systems, can increase the speed and scale at which harmful biases manifest and are acted upon”
- NIST AI 100-2 E2025, Adversarial Machine Learning glossary: training stage ↗ “The stage of a machine learning pipeline in which a model learns parameters that minimize its error against an objective function based on training data.”
- UNESCO: Generative AI, UNESCO study reveals alarming evidence of regressive gender stereotypes (7 March 2024) ↗ “a UNESCO study revealed worrying tendencies in Large Language models (LLM) to produce gender bias, as well as homophobia and racial stereotyping.”
Question and answer: UPSC's provisional GS Paper I (2026, Series A) — paper ↗ · answer key ↗. Explanation: Minimalist IAS, checked 30 Sept 2026 (how we verify). ·