Dave Rahardja on Nostr: #AI #LLM is gunking up the web, especially for lesser-represented languages. Spammers ...
#AI #LLM is gunking up the web, especially for lesser-represented languages. Spammers are creating garbage English language content using LLMs, then translating it into *multiple languages* at the same time, using Machine Translation, presumably to generate clickbait ad revenue in several languages at once.
In English, such gunk accounts for some 9% of total sampled web content. But in languages with less representation on the Internet, the figures could be much higher. In Malay, it’s something like 26%, and in Swahili it’s nearly HALF of everything found on the web.
Paper [pdf]: “A Shocking Amount of the Web is Machine Translated: Insights from Multi-Way Parallelism”
https://arxiv.org/pdf/2401.05749.pdfPublished at
2024-01-25 02:54:07Event JSON
{
"id": "b222944fef275131031e671c848d1459718215f52b5aa864d0e955e19092168f",
"pubkey": "8ca0240eaf6bc332736f59eb486087e2113d7b599910cea6bd90b6a21460134f",
"created_at": 1706151247,
"kind": 1,
"tags": [
[
"t",
"ai"
],
[
"t",
"LLM"
],
[
"proxy",
"https://sfba.social/users/drahardja/statuses/111814328155829448",
"activitypub"
]
],
"content": "#AI #LLM is gunking up the web, especially for lesser-represented languages. Spammers are creating garbage English language content using LLMs, then translating it into *multiple languages* at the same time, using Machine Translation, presumably to generate clickbait ad revenue in several languages at once.\n\nIn English, such gunk accounts for some 9% of total sampled web content. But in languages with less representation on the Internet, the figures could be much higher. In Malay, it’s something like 26%, and in Swahili it’s nearly HALF of everything found on the web.\n\nPaper [pdf]: “A Shocking Amount of the Web is Machine Translated: Insights from Multi-Way Parallelism”\n\nhttps://arxiv.org/pdf/2401.05749.pdf",
"sig": "f204bf66f9f27baee19d4b55c6b55ada6403c1022ad84152a726085f5e9b2ebf9a6a59df6d91a5bb04c35e111896975d1525992ed0a2a20e532715261e04a224"
}