What are the use cases for decoder-based transformers?

• A. Text generation, chatbots, and code generation.
• B. Text classification, semantic search, and NER.
• C. Image captioning, object detection, and style transfer.
• D. Speech synthesis, translation, and summarization.

Answer: (A)

Decoder-based transformers shine at generating text because they are trained to predict the next token given all prior tokens, an autoregressive setup that makes fluent, coherent generation their strength. This naturally maps to use cases like text generation, where the model continues a prompt or creates long passages in a chosen style; chatbots, where it must produce plausible, context-aware replies in a conversation; and code generation, where it writes source code that follows syntax rules and programmers’ intent.

Other tasks rely more on understanding input and producing specific outputs or transferring information across modalities. Text classification, NER, and similar labeling tasks depend on creating and applying representations to decide labels, rather than continuing text. Image captioning or other multimodal tasks require an image understanding component plus a generation step, often using an encoder to process the image before generation. Translation and summarization can be done by decoder-based models, but they’re more typically handled with encoder–decoder setups that map a source sequence to a target sequence, leveraging explicit input understanding before generating output.