Quick question. Do you talk to your AI in English or your native language?

I’ve started treating input language as context. Not just preference.

Recipe for “bolos de bacalhau”? Portuguese. Code review? English. Visa requirements for a non-English speaking country? I translate my question first, ask in that language, then translate the answer back myself. Sounds paranoid. Maybe is. But the results feel better.

Here’s my theory. When you ask a question in English about something deeply regional, the AI pulls from English-language sources. Which might be fine. Or might be some tourist blog from 2019 that got it wrong.

Ask the same question in the local language and it’s drawing from different training data. Local forums. Government sites. Actual native speakers writing about their own stuff.

Same with idiomatic expressions. Ask in English, get a translation. Ask in the original language, get the real phrase people actually use.

The other thing: mixing languages in one prompt seems to cost something. The model is reasoning and translating at the same time. More moving parts, more room for drift.

Could be I’m pattern-matching on noise. But it feels real enough that I keep doing it. Anyone else doing this? Or am I just hallucinating like an AI?

Media

⚠️ Media not available locally