BitbyBit on Nostr: LLMs often mess up spelling in images because they read words as chunks and not ...
LLMs often mess up spelling in images because they read words as chunks and not letters.
That’s why AI-generated text can look weird or wrong as these models just aren’t built for letter-by-letter detail.
For me, Grok has been the most accurate with text on images so far.
Published at
2025-06-10 09:05:25Event JSON
{
"id": "383e6a4c467c695c8e78d3513d71adf4c242578cb3f0f8cdd2eda0f53edd330a",
"pubkey": "fbdd6e1de892ca9c6874810bd90a569de1a0b80955f558ba15bfe3d73cf87d7d",
"created_at": 1749546325,
"kind": 1,
"tags": [],
"content": "LLMs often mess up spelling in images because they read words as chunks and not letters.\n\nThat’s why AI-generated text can look weird or wrong as these models just aren’t built for letter-by-letter detail.\n\nFor me, Grok has been the most accurate with text on images so far.",
"sig": "e391b1774f1e312d7d3e9672d630898eb2547c23247c0bab8e2d90bbff6f99c209bc950556ae7abf9e85fb705397346444f5e62ec68ed01876e4b13d8d4c246d"
}
