Which neural network architecture is primarily used by LLMs?

Question

Options

• ATransformer architecture with self-attention mechanisms
• BRecurrent neural network architecture
• CConvolutional Neural Networks (CNNs)
• DSequential encoder-decoder architecture

• A
  93% (26)
• C
  4% (1)
• D
  4% (1)

Explanation

Large Language Models are built on the Transformer architecture, which uses self-attention mechanisms to weigh relationships between all tokens in a sequence simultaneously - enabling the model to capture long-range dependencies far more effectively than older designs. Recurrent Neural Networks (B) process tokens sequentially, making them slow to train and poor at retaining context over long sequences, which is why they were largely replaced by Transformers. CNNs (C) are optimized for spatial pattern recognition in grid-like data (images), not for modeling sequential language. The sequential encoder-decoder (D) is actually a component pattern that early seq2seq models used - but modern LLMs like GPT are decoder-only Transformers, not traditional encoder-decoder pipelines.

Memory tip: Think of the "T" in GPT - it stands for Transformer. Every major LLM family (GPT, BERT, LLaMA, Claude) is Transformer-based, so when in doubt, the answer involving self-attention is almost always correct.

No community discussion yet for this question.