cd /news/large-language-models/ask-hn-can-you-still-tell-ai-generat… · home topics large-language-models article
[ARTICLE · art-100534] src=news.ycombinator.com ↗ pub= topic=large-language-models verified=true sentiment=· neutral

Ask HN: Can you still tell AI-generated text apart in your own language?

A Japanese user on Hacker News reports that AI-generated Japanese text remains detectable due to unnatural word choices such as 実務 and 帳簿, while AI-generated English has improved significantly. The user notes that model choice matters more outside English, citing Anthropic's Opus 4.6 and Google's models as producing better Japanese, whereas OpenAI's models and Opus 4.8/5 sound too geeky.

read1 min views1 publishedAug 17, 2026

I'm Japanese, and I can still usually spot AI-generated Japanese by its word choice. Some words, like 実務 (practical work) and 帳簿 (a ledger), turn up far more often than they should. Casual ones like 効く (to work well, to be effective) and 刺さる (to pierce > to resonate, to hit home) get used where another word would normally be the first choice. They're the Japanese equivalent of "delve."

I'm not a native English speaker, but my impression is that AI-generated English has improved a lot recently and is harder to pick out than it used to be. Does that match what English native speakers see?

What about other languages? Chinese must be one of the most heavily trained languages, so I'd expect it to have improved as much as English. And what about Spanish, Hindi, and other widely spoken languages — can speakers still tell?

Also, I think which model you use still matters far more once you leave English. One of my first singularity moments came from Gmail's AI drafting feature. It had clearly absorbed the conventions of Japanese corporate email culture, bureaucracy and all.

These days, Anthropic's Opus 4.6 (and Fable 5 outside coding sessions) writes reasonably good long-form Japanese, and Google's models hold up across a wider range of registers. OpenAI's models, Opus 4.8/5 and Fable in a coding session sound way too geeky.

Curious to hear your thoughts.

Comments URL: [https://news.ycombinator.com/item?id=49338865](https://news.ycombinator.com/item?id=49338865)

Points: 1

── more in #large-language-models 4 stories · sorted by recency
── more on @hacker news 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/ask-hn-can-you-still…] indexed:0 read:1min 2026-08-17 ·