Feature
Multilingual AI disclaimer rewriting
Dropping a hedge is a grammar problem, not a universal find-and-delete operation.
Why translation is not enough
The same English idea appears as a conjugation, modal verb, suffix, or whole clause in other languages. Removing one token can make the sentence ungrammatical or change the subject agreement.
Makes Mistakes keeps an explicit pattern and replacement for each phrasing it supports. It replaces only the hedge and preserves the model name and surrounding interface text.
Examples of the grammatical problem
| Language pattern | Why it needs its own rule |
|---|---|
| English: can make mistakes | The modal and verb phrase become a third-person singular statement |
| German: kann Fehler machen | The finite verb and infinitive structure must be rearranged |
| Hungarian: hibázhat | The possibility is encoded in the verb suffix |
| Japanese and Korean disclaimer clauses | The hedge can span a longer formal phrase rather than one word |
| Right-to-left scripts | Phrase matching and inserted punctuation or emoji must preserve readable order |
Coverage and verification limits
The repository contains rules across dozens of language families and scripts, based on wording observed in live interfaces. The project welcomes native-speaker corrections because grammatically plausible is not the same as naturally phrased.
Provider copy and DOM structure can change independently. A language can remain supported while a new disclaimer wording needs an additional pattern.
What the extension does not translate
- It does not translate your conversation.
- It does not change the provider's general interface language.
- It does not translate the fixed English Improve answer nudge automatically.
- It does not claim every rule has been reviewed by a native speaker.
FAQ
Questions people ask
How many languages does Makes Mistakes support?
The code includes phrase rules across dozens of languages and scripts. Because providers can change their wording and some rules still welcome native review, the project describes the coverage qualitatively rather than promising a permanent exact count.
Does it detect my browser language?
It responds to the actual disclaimer text rendered by the AI service, so the matching rule follows the interface wording on the page.
Can I suggest a better translation?
Yes. The project is open source, and language corrections can be proposed with the observed provider wording and the more natural direct form.