So basically that is what I was saying. Imagine an LLM that can also OCR. Would it be possible to make it OCR a totally new letter by only showing a single pic

So basically that is what I was saying. Imagine an LLM that can also OCR. Would it be possible to make it OCR a totally new letter by only showing a single picture of it and including the fact in the context? I think it would not be possible. That would be a good demonstration of the point I (and possibly you as well) is trying to get across.