Text gets mysteriously reversed in AI Summarization payload

Fout
1

Most of my summaries are failing, because some of the text sent to the LLM is reversed(!). If the post is simply:

This post is to test whether sections of text get reversed in the payload sent to the AI endpoint.

it actually sends this text to the LLM:

This post is to test whether sections of text get .tniopdne IA eht ot tnes daolyap eht ni desrever

In this simple test, it folds or reflects exactly at the middle of the string.

I do realize this sounds crazy, so here is a screenshot and the corresponding raw_request_payload from the ai_api_audit_logs:

image
image1022×512 45 KB

ai_api_audit_logs.raw_request_payload 
{
  "model": "claude-sonnet-4-0",
  "max_tokens": 64000,
  "messages": [
    {
      "role": "user",
      "content": "Here are the posts inside <input></input> XML tags:\n\n<input>1) user1 said: I love Mondays 2) user2 said: I hate Mondays</input>\n\nGenerate a concise, coherent summary of the text above maintaining the original language."
    },
    {
      "role": "assistant",
      "content": "{\"summary\":\"Two users are sharing their feelings toward Mondays. [user1](/t/-/64255/1) hates them, while [user2](/t/-/64255/2) loves them.\"}"
    },
    {
      "role": "user",
      "content": "The discussion title is: Testing Text Reversal.\n\nCategory: testing.\n\n\nHere are the posts, inside <input></input> XML tags:\n\n<input>\n  (1 admin said: This post is to test whether sections of text get .tniopdne IA eht ot tnes daolyap eht ni desrever  \n</input>\n\nGenerate a concise, coherent summary of the text above maintaining the original language."
    },
    {
      "role": "assistant",
      "content": "{"
    }
  ],
  "system": "You are an advanced summarization bot that generates concise, coherent summaries of provided text.\nYou are also capable of enhancing an existing summaries by incorporating additional posts if asked to.\n\n- Only include the summary, without any additional commentary.\n- You understand and generate Discourse forum Markdown; including links, _italics_, **bold**.\n- Maintain the original language of the text being summarized.\n- Aim for summaries to be 400 words or less.\n- Each post is formatted as \"<POST_NUMBER>) <USERNAME> <MESSAGE>\"\n- Cite specific noteworthy posts using the format [DESCRIPTION](/t/-/64255/POST_NUMBER)\n- Example: link to the 6th post by jane: [agreed with](/t/-/64255/6)\n- Example: link to the 13th post by joe: [joe](/t/-/64255/13)\n- When formatting usernames use [USERNAME](/t/-/64255/POST_NUMBER)\n\nFormat your response as a JSON object with a single key named \"summary\", which has the summary as the value.\nYour output should be in the following format:\n\n{\"summary\": \"xx\"}\n\nWhere \"xx\" is replaced by the summary.\nreply with valid JSON only",
  "stream": true
}

Interestingly, Sonnet refuses to summarize many topics, because it thinks the reversed text is an effort to obfuscate or get around the rules.

I’m up to date at 3.6.0.beta3-latest.

1 like