Orbital

Loading this week's intelligence

Pulling ranked themes, signals, and content outputs from the API.

Overview Operator registry

Health posturesucceeded

Last checked 03 Sept 2026, 06:25 am

Change postureunknown

Last changed 03 Sept 2026, 06:25 am · next check 10 Sept 2026, 06:25 am

Registry postureEnabled

youtube.com

Lifecycle posturecandidate

No receipt yet

Metadata

Operator-facing source metadata and safe config fields.

Slug: yt-yplW8dHsxqI-d9654309
Domain: youtube.com
Jurisdiction: n/a
Layer: n/a
Cadence: weekly
Change state: n/a
Content label: n/a
Feed URL: n/a
Base URL: https://www.youtube.com/watch?v=yplW8dHsxqI
Fetch method: n/a
Fetch strategy: n/a
Last checked: 03 Sept 2026, 06:25 am
Last changed: 03 Sept 2026, 06:25 am
Next check: 10 Sept 2026, 06:25 am
Cache posture: n/a

No tags

Recent ingestion history

Most recent source-specific ingestion runs and operator-visible failure context.

Recent documents

Latest normalized output from this source.

https://www.youtube.com/watch?v=yplW8dHsxqI

Recent errors

Last visible failure reasons to help operators diagnose source health quickly.

No recent errors

Recent failure reasons will appear here when runs fail.

Config JSON

{
  "title": "Meta-Prompting: Ask the LLM to Write Your Prompt for Better Results.",
  "video_id": "yplW8dHsxqI",
  "channel_id": "UCEWLCMe8iwKOakYoaQ2k1eg",
  "fetched_at": "2026-08-13T06:20:50.363479+00:00",
  "description": "Meta-prompting is a prompt engineering technique where a language model is asked to write the prompt itself, instead of the user guessing at the perfect wording. Rather than requesting an answer directly, the request is raised one level: the model produces a detailed prompt that covers everything that matters for the task. One card asks for an answer; the other asks for a better question.\n\nWhy it matters: an under-specified prompt forces the model to fill the gaps with bland defaults. A request like 'write a birthday invitation' returns a flat result with no date, no place, and no warmth, and the user cannot even tell what is missing. A meta-prompt instead returns a candidate prompt, often a checklist asking for the name and age, the date, time, place, theme, and RSVP — questions the user never thought to ask.\n\nThe core flow:\n- A goal feeds into a meta-prompt.\n- The model writes a candidate prompt as output.\n- The candidate prompt is run with real details, producing a complete answer.\n",
  "topic_seeds": [
    "Underspecified Machine Learning Theme"
  ],
  "caption_kind": "asr_auto",
  "channel_name": "Alfa Yohannis",
  "published_at": "2026-07-25T00:38:41Z",
  "source_universe": {
    "cadence": "weekly",
    "warnings": [],
    "feed_urls": [],
    "source_type": "youtube_video",
    "sitemap_urls": [],
    "allowed_paths": [],
    "blocked_paths": [],
    "crawl_posture": "manual_approval",
    "domain_family": "youtube.com",
    "freshness_sla": "weekly",
    "source_family": "video_media",
    "trust_posture": "community",
    "authority_tier": "tier_d",
    "branch_relevance": [],
    "historical_yield": {},
    "rejection_history": [],
    "source_family_raw": "video_media",
    "audience_relevance": [],
    "contamination_history": [],
    "robots_unavailable_policy": "fail_closed"
  },
  "transcript_text": "Welcome to metaprompting. The idea is\nactually simple. Instead of guessing the\nperfect prompt, ask the model itself to\nwrite it for you. You get far richer,\nmore complete results.\nA parent wants a birthday invitation for\ntheir daughter. They type just one short\nline, write a birthday invitation. Back\ncomes three flat lines, no date, no\nplace, no warmth. When a prompt is this\nunderspecified, the model fills the gaps\nwith bland defaults, and the parent\ncan't even tell what's missing.\nHere's the key move. Raise your prompt\none level. Don't ask for the invitation\ndirectly. Instead, ask the model to\nwrite a detailed prompt covering\neverything that matters. See the\ndifference? One card wants an answer.\nThe other wants a better question.\nThe flow is simple. Your goal feeds into\nthe metaprompt, and the model writes a\ncandidate prompt as output. Remember,\nnothing is trained. It's all just text\nin the context window.\nNow run that candidate prompt with your\nreal details, and a complete answer\nemerges. Not satisfied with the result?\nAsk the model to critique and rewrite\nthe prompt, then repeat.\nThe metaprompt then returns a candidate\nprompt, a simple checklist covering the\nchild's name and age, the date, time,\nplace, theme, and RSVP. Questions the\nparent never thought to ask.\nNow fill each blank with real details,\nname, age, theme, time, and place. Then\nrun the prompt, and back comes a warm,\ncomplete invitation. Compared with the\nflat first draft, the quality jump is\ninstantly clear.\nWhen does metaprompting help? If your\nresults keep coming back bland, or you\nneed a template you can reuse, it\nsupplies the structure you've been\nmissing. But for a simple one-off task,\nthe extra round trip just adds cost. And\nremember, generated prompts can carry\ntheir own blind spots, so review them\nyourself.\nIn short, ask the model to write the\nprompt for you. Run that prompt with\nyour real details, and refine it once if\nneeded. The flow, goal, generated\nprompt, then answer.",
  "duration_seconds": 134,
  "is_auto_generated": true,
  "transcript_source": "youtube_transcript_api",
  "transcript_status": "available",
  "asr_quality_weight": 0.6,
  "transcript_segments": [
    {
      "text": "Welcome to metaprompting. The idea is",
      "start": 0,
      "duration": 5
    },
    {
      "text": "actually simple. Instead of guessing the",
      "start": 2.96,
      "duration": 4.56
    },
    {
      "text": "perfect prompt, ask the model itself to",
      "start": 5,
      "duration": 4.84
    },
    {
      "text": "write it for you. You get far richer,",
      "start": 7.52,
      "duration": 4.84
    },
    {
      "text": "more complete results.",
      "start": 9.84,
      "duration": 4.44
    },
    {
      "text": "A parent wants a birthday invitation for",
      "start": 12.36,
      "duration": 4.28
    },
    {
      "text": "their daughter. They type just one short",
      "start": 14.28,
      "duration": 4.92
    },
    {
      "text": "line, write a birthday invitation. Back",
      "start": 16.64,
      "duration": 5.4
    },
    {
      "text": "comes three flat lines, no date, no",
      "start": 19.2,
      "duration": 5.72
    },
    {
      "text": "place, no warmth. When a prompt is this",
      "start": 22.04,
      "duration": 5.28
    },
    {
      "text": "underspecified, the model fills the gaps",
      "start": 24.92,
      "duration": 4.36
    },
    {
      "text": "with bland defaults, and the parent",
      "start": 27.32,
      "duration": 4.84
    },
    {
      "text": "can't even tell what's missing.",
      "start": 29.28,
      "duration": 5.08
    },
    {
      "text": "Here's the key move. Raise your prompt",
      "start": 32.16,
      "duration": 4.44
    },
    {
      "text": "one level. Don't ask for the invitation",
      "start": 34.36,
      "duration": 4.96
    },
    {
      "text": "directly. Instead, ask the model to",
      "start": 36.6,
      "duration": 4.56
    },
    {
      "text": "write a detailed prompt covering",
      "start": 39.32,
      "duration": 3.76
    },
    {
      "text": "everything that matters. See the",
      "start": 41.16,
      "duration": 4.44
    },
    {
      "text": "difference? One card wants an answer.",
      "start": 43.08,
      "duration": 5.28
    },
    {
      "text": "The other wants a better question.",
      "start": 45.6,
      "duration": 5.28
    },
    {
      "text": "The flow is simple. Your goal feeds into",
      "start": 48.36,
      "duration": 4.4
    },
    {
      "text": "the metaprompt, and the model writes a",
      "start": 50.88,
      "duration": 4.64
    },
    {
      "text": "candidate prompt as output. Remember,",
      "start": 52.76,
      "duration": 5.24
    },
    {
      "text": "nothing is trained. It's all just text",
      "start": 55.52,
      "duration": 4.96
    },
    {
      "text": "in the context window.",
      "start": 58,
      "duration": 4.52
    },
    {
      "text": "Now run that candidate prompt with your",
      "start": 60.48,
      "duration": 4.12
    },
    {
      "text": "real details, and a complete answer",
      "start": 62.52,
      "duration": 4.8
    },
    {
      "text": "emerges. Not satisfied with the result?",
      "start": 64.6,
      "duration": 4.88
    },
    {
      "text": "Ask the model to critique and rewrite",
      "start": 67.32,
      "duration": 5.16
    },
    {
      "text": "the prompt, then repeat.",
      "start": 69.48,
      "duration": 5.04
    },
    {
      "text": "The metaprompt then returns a candidate",
      "start": 72.48,
      "duration": 4.32
    },
    {
      "text": "prompt, a simple checklist covering the",
      "start": 74.52,
      "duration": 5.56
    },
    {
      "text": "child's name and age, the date, time,",
      "start": 76.8,
      "duration": 6.68
    },
    {
      "text": "place, theme, and RSVP. Questions the",
      "start": 80.08,
      "duration": 6.16
    },
    {
      "text": "parent never thought to ask.",
      "start": 83.48,
      "duration": 5.36
    },
    {
      "text": "Now fill each blank with real details,",
      "start": 86.24,
      "duration": 6.84
    },
    {
      "text": "name, age, theme, time, and place. Then",
      "start": 88.84,
      "duration": 6.76
    },
    {
      "text": "run the prompt, and back comes a warm,",
      "start": 93.08,
      "duration": 4.52
    },
    {
      "text": "complete invitation. Compared with the",
      "start": 95.6,
      "duration": 4.68
    },
    {
      "text": "flat first draft, the quality jump is",
      "start": 97.6,
      "duration": 4.96
    },
    {
      "text": "instantly clear.",
      "start": 100.28,
      "duration": 4.32
    },
    {
      "text": "When does metaprompting help? If your",
      "start": 102.56,
      "duration": 4.12
    },
    {
      "text": "results keep coming back bland, or you",
      "start": 104.6,
      "duration": 4.08
    },
    {
      "text": "need a template you can reuse, it",
      "start": 106.68,
      "duration": 3.32
    },
    {
      "text": "supplies the structure you've been",
      "start": 108.68,
      "duration": 4
    },
    {
      "text": "missing. But for a simple one-off task,",
      "start": 110,
      "duration": 5.44
    },
    {
      "text": "the extra round trip just adds cost. And",
      "start": 112.68,
      "duration": 4.88
    },
    {
      "text": "remember, generated prompts can carry",
      "start": 115.44,
      "duration": 4.52
    },
    {
      "text": "their own blind spots, so review them",
      "start": 117.56,
      "duration": 4.4
    },
    {
      "text": "yourself.",
      "start": 119.96,
      "duration": 4.24
    },
    {
      "text": "In short, ask the model to write the",
      "start": 121.96,
      "duration": 4.12
    },
    {
      "text": "prompt for you. Run that prompt with",
      "start": 124.2,
      "duration": 4.199
    },
    {
      "text": "your real details, and refine it once if",
      "start": 126.08,
      "duration": 5.24
    },
    {
      "text": "needed. The flow, goal, generated",
      "start": 128.399,
      "duration": 6.241
    },
    {
      "text": "prompt, then answer.",
      "start": 131.32,
      "duration": 3.32
    }
  ],
  "view_count_at_fetch": 9,
  "transcript_available": true,
  "source_universe_backfill": {
    "version": "source_universe_backfill.v1",
    "backfilled_at": "2026-08-15T06:03:50.119874+00:00",
    "resolver_warnings": [],
    "resolved_source_family": "video_media"
  },
  "transcript_last_attempted_at": "2026-08-13T06:20:58.204660+00:00"
}