Decoder. plain-English AI glossary

Prompt compression

▲ Rising

Reducing a long prompt or context to its essential meaning — preserving information while shrinking token count.

Think of it like

Summarizing a 50-page report into a 5-page executive summary.

Example

A 500-token document compressed to 200 tokens via summarization or selector techniques, then used in a prompt.

How it actually works

Compression can be lossy (summary) or lossless (clever encoding). Lossy is simpler but risks missing details. Lossless is hard (compression algorithms don't beat LLMs at understanding semantics). Some models specifically trained for compression exist. Metrics: information retention vs. token savings.

For product teams

Reduces cost for long-context tasks.

For engineers

Experiment with summarization + compression. Evaluate retention with eval set.

Related

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open DecoderAdd to Chrome