I've noticed a huge negative shift since services like google translate started moving over from simple algorithms to LLM-based tech. In the past I could stick a paragraph into google translate and get a result that was clunky but at least a literal translation. +
cypherstate⁷
yeah, that's what's really insidious about deepL and other LLM-based translation tools -- they're skewed to favor output that "looks right" even if it's pure hallucination, and someone who can't read the source text (AKA everybody, since why else would you need a translator) is going to be clueless
