There are no lossless transformations of natural-language text
The fundamental problem nobody's talking about: every text transformation an AI makes loses information.
Why it matters
A technical deep-dive on why lossless transformations of natural language are theoretically impossible — a constraint that affects everything from context windows to reasoning chains to agent reliability.
The key facts
10 to knowNo lossless transformation exists for natural-language text
Information loss is inherent to any compression, summarization, or reformatting
Implications for context-window management, token efficiency, and reasoning chain fidelity
Published by Simon Willison (noted AI researcher and Datasette creator)
Fundamental computer science / information theory argument, not vendor-specific
Core claim: no lossless transformations exist for natural-language text
Source: Simon Willison (trusted ML commentator and Datasette creator)
Published Aug 11, 2026 — contemporary research finding
Directly impacts tokenization strategies, context efficiency, and information loss in LLM pipelines
Theoretical insight into model architecture constraints
Go to the source
Simon Willisonsimonwillison.net