{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreidcqjiv7oipup4vwjq4kbvcl4cjvdbvszbr4edh3vq624ouk222ce",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mk4azprwndm2"
  },
  "path": "/t/repeat-phrase-instructions-broken-in-gpt-realtime-1-5/1379147#post_4",
  "publishedAt": "2026-04-22T18:17:08.000Z",
  "site": "https://community.openai.com",
  "tags": [
    "@dbhc"
  ],
  "textContent": "Hi @dbhc, yeah this one’s a bit jarring if you’re coming from the older model behavior.\n\nThe shift is real. The newer realtime model treats instructions more like guidance than strict rules, so things that used to act like hard constraints now get interpreted more loosely.\n\nFor example:\n\n**Older model:**\n\nWhen you can’t clearly hear the user, don’t proceed. If there’s background noise or you only caught part of the sentence, pause and ask them politely to repeat themselves in their preferred language, and keep the conversation in the same language.\n\n**Newer model:**\n\nOnly respond to clear audio or text.\n\nIf audio is unclear/partial/noisy/silent, ask for clarification in {preferred_language}.\n\nContinue in the same language as the user if intelligible.\n\nThe newer version works better because it’s broken into short, explicit rules instead of one long instruction. That structure tends to stick more reliably with 1.5.\n\nAlso, you might find the “Using realtime models” prompting guide helpful since it’s aligned with how the newer model behaves.\n\nCurious if tightening the prompt like this improves things on your side.\n\n-Mark G.",
  "title": "Repeat phrase instructions broken in gpt-realtime-1.5"
}