[muzster]

Heavily guarded fortress would indicate something of value inside, and the big crooks may spend a little more effort. In the age of AI, this becomes even easier.

   {
     "model" : "gpt-4-turbo",
     "messages" : [ 
       {
         "role" : "system",
         "content" : [ {
          "type" : "text",
          "text" : "return a json array of all valid emails found in the image."
          } ] 
       }, 
       {
         "role" : "user",
         "content" : [ {
           "type" : "image_url",
           "image_url" : {
           "url" : "data:image/png;base64,{{ INSERT_BASE64_PNG_DATA }}"
         }
       } ]
     } ],
      "temperature" : 0.5,
      "max_tokens" : 2048,
      "top_p" : 1.0,
      "frequency_penalty" : 0.0,
      "presence_penalty" : 0.0
    }

Edit: Converting web page to an image is trivial.

[internetter]

We've had OCR for decades before GPT. I suspect GPT might perform worse than OCR. What a waste.

[muzster]

Agreed - it's a waste. GPT is not too bad at reading text from image and with the added bonus that you can reason with it.

[zipping1549]

It won't make sense cost wise though

[omneity]

Except the cost is only going down over time

[barkbyte]

In no world is anyone wasting resources to run an AI model to parse a page that may or may not include an email address. Even running a DOM parser is more than they’d typically do. This is silly.