Do spelling mistakes and abbreviations in prompts help or hurt response quality? Tokenizer but also meaning
retained text
If you spell "Let's" as "LEt's" or refer to github (not the CLI) as GH or you say "teh" instead of "The", the tokenizer takes another token with another set of correlations. But does that extra attention potentially help. or is the harness correcting a lot of these anyway?
document fetched 2026-10-09 00:43 · canonical https://reddit.com/r/ClaudeCode/comments/1x12ud5/do_spelling_mistakes_and_abbreviations_in_prompts