Trust Real Intelligence, Not the Artificial Kind

Artificial intelligence can translate your documents in seconds, taking human translators out of the loop and saving you time and money.

But is it true?

Sure, AI can speed up repetitive tasks, and Generative AI can appear to create new content for you, but its smooth language and pretty-sounding text covers up mistakes you may not even see without an expert eye watching out for you.

If you don’t speak a foreign language and ask AI to translate a document for you, how do you know the machine got it right? I decided to find out.

I chose two publicly available documents: a patient education page from the Argentine government on managing celiac disease, and a peer-reviewed journal article about a patient with both celiac disease and epilepsy. I uploaded both documents to several large language models (LLMs) that offer free accounts and asked them to translate the documents. I then compared the translations side-by-side with the originals.

The results were sometimes chuckle-worthy, often concerning, and across-the-board illuminating. Here’s a preview of what I found:

• The AI translation often sounds great – so great that it hides the problems it creates.
• Yes, creates – all four LLMs I tested made mistakes a human translator and editor wouldn’t dream of making.
• The general idea of the text came across more or less, but…
• The AI tools often changed the structure and formatting of the article, and sometimes it got things wrong.
• Every single LLM even forgot to translate parts of the text!

If you’d like to know

• Which tools decided baking ingredients might be emotionally unstable
• Which tool decided gluten-free flour is perfectly safe in the kitchen but shouldn’t be taken outdoors
• Which LLM can’t help but add bullet points to everything
• How much information each tool forgot to translate
• Which LLM wrote extra text that didn’t appear on the original page

Just send me a message to get the full report.