Most people assume that the typos can confuse the model. When I first started using ChatGPT Gemini, or Claude, I constantly caught myself fixing misspelled words before I was hitting enter. I felt that the prompt should be perfect before I submitted to AI. And that leads to the natural question. Does correcting typos actually matter? And especially for long texts, should you really change the misspelled words and fix every mistake? Here is the important part. The text is not cleaned and it's not corrected before it's submitted for processing.
The model sees your text exactly as you typed it. That's because the model doesn't read words, it reads tokens, small chunks of characters. So when you write "unreasonable" and you make a mistake in the word "unreasonable", then most of the tokens still the same and nothing is really changing. The shape is similar, the pattern survives. And because the meaning in the language is highly redundant, the continuation still points to the same idea. Typos don't disappear, they are still there, they just usually don't matter that much.
This works because models were trained on a messy text, the internet is full of broken grammar, misspelling and mixed languages. So the model isn't fragile, but here is the limit. If a typo changes the word into a different one, or the meaning changes completely, then it may matter. And if a typo breaks structure like code, formulas, IDs or passwords, then the meaning collapses. Because now the pattern is gone. So the main takeaway is simple. AI doesn't understand spelling, it understands patterns. And as long as the pattern survives, as long as the word is still the same, typos usually don't matter.
This was a Note on AI for today, thank you for watching and I'll see you in the next one.
