Andrej Karpathy — Intro to LLMsY Combinator — How to get startup ideas
Your YouTube video
Paste a YouTube link. Nothing to upload.
←YouTube video
Make a blog post
Turn a YouTube video into a blog post
Paste a link. Ancher reads the captions and the frames, then drafts an article you can actually publish.
No link handy? Try one:
Free to try, no card. Your YouTube video opens in Ancher.
Caption track with timingsKey frames from the videoDescription and linked sources
Kept in parts
Every one of these stays separately addressable instead of collapsing into one block of text.
The angleOpeningBody sectionsSpecifics and numbers
Blog post
A publishable article with its own argument, not a transcript rearranged into paragraphs.
Ancher reads the video's caption track and samples frames from the video itself, so the numbers a creator put on a slide but never said out loud make it into the draft. The article is then built around the claim the speaker defends hardest rather than around the video's running order, and every specific keeps a timecode you can jump back to before you publish.
The honest version
Transcript-only tools lose the part that makes a post worth reading
Nearly every video-to-blog tool works from the caption track alone. That is fine for the argument and useless for the evidence, because on YouTube the specifics live on screen — a pricing table, a benchmark chart, a screenshot of a dashboard. A caption scrape sees none of it, so the draft comes back fluent and empty, and you go back to scrubbing the video for the numbers yourself.
One second of one video, at 14:07, read two ways.Illustrative. The figure is invented; the gap it falls into is not.
...and that brought our cost per lead down to which is why we kept the channel running into Q4.
Cost per lead, Q3 — not readThe specific never reaches the draft
Blog post
Cost per lead, Q3
$4.20
from 14:07
Lands in the draft, with its timecode
- Ancher samples and scores frames from the video, so on-screen figures reach the draft as evidence rather than being lost at ingest.
- Every extracted claim keeps its timecode, so fact-checking a number is one click rather than a rewatch.
- The creator's description links come along, so the finished post can cite the primary source instead of citing the video.
What comes out
Blog post, section by section
A publishable article with its own argument, not a transcript rearranged into paragraphs.
The angle
One sentence, taken from the claim the speaker spends the most time defending — not from the video title, which was written for a thumbnail.
Opening
Opens on the strongest specific, even if it arrived at minute forty. A written piece front-loads what a video saves for the end.
Body sections
Sections reordered for a reader rather than a viewer, with the video's own chaptering treated as raw material, not as an outline.
Specifics and numbers
Every number, date and named entity from the video, each tagged with whether it was spoken, shown on screen, or both.
Sources and credit
The creator credited and linked, plus the sources they themselves cited, so the post can stand on primaries.
YouTube-to-blog-post prompt
The link on its own gets a summary. This is the instruction that produces the 5 sections above, in that order, with the rules that keep them honest. Paste it with your YouTube video — in Ancher, or in whatever assistant you already use.
From this video, write a blog post. (1) State the angle in one sentence, drawn from the claim the speaker defends at greatest length — not from the video title. (2) Open on the single most specific moment in the video, wherever it occurred. (3) Order the sections for a reader, conclusion first; do not follow the video's chronology. (4) List every number, date and named entity, and tag each as SPOKEN, ON-SCREEN, or BOTH, with its timecode. (5) Credit the creator and link the video, and cite the description's sources rather than the video wherever the video was quoting someone else. Flag anything you could not verify against a frame or a caption line instead of smoothing it over.
The detail, if you want it
How each part of a YouTube video becomes a section
| From the YouTube video | Into | Why |
|---|---|---|
| Caption track with timings | The angle | The angle is derived from where the speaker spends time defending a position, which the caption track shows and the video title actively hides. |
| Key frames from the video | Specifics and numbers | Frames are where the pricing table and the benchmark chart live; this is the row every caption-only competitor cannot fill. |
| Caption track with timings | Body sections | Segment timings expose where the video changes subject, which gives a section break that a reader will accept. |
| Description and linked sources | Sources and credit | Description links let the finished post cite what the creator cited, instead of citing a video for a claim the video borrowed. |
| Likes and comment counts | Opening | Comment volume on a video is a blunt but real signal about which claim readers actually argued over, which is usually the right opening. |
Everything Ancher keeps from a YouTube video
Caption track with timings
Published and auto-generated captions both come through, and every line keeps the second it was spoken.
Key frames from the video
Frames are sampled and scored, so a number shown on a slide survives even though it was never said out loud.
Description and linked sources
The creator's own cited links come along, so a claim can be traced past the video.
Likes and comment counts
Reception is a weak signal about correctness and a strong one about which claims got attention.
Title, channel, duration, date
Basic provenance, kept so the finished work can attribute the video properly.
The four steps
- Add the YouTube video. Paste a YouTube URL. Ancher pulls the caption track with its timings, samples key frames from the video itself, and keeps the description links and engagement counts on the same source.
- Nothing gets flattened. Caption track with timings, key frames from the video, description and linked sources, likes and comment counts, title, channel, duration, date stay separately addressable instead of collapsing into one block of text.
- Ask for a blog post. Ancher fills the angle, opening, body sections, specifics and numbers, sources and credit using the mapping above, not a generic summary pass.
- Check it before you use it. Every specific keeps a link back to where it came from in the YouTube video, so a claim can be verified without reopening the source.
Where this goes wrong
- Auto-generated captions mangle spoken figures — fifteen and fifty collapse constantly. Any number that matters should be checked against the frame it appeared in before the post goes out.
- A draft that follows the video's running order reads as a transcript in paragraphs. The test is whether the opening line would survive if the video were deleted.
- Sponsored segments transcribe exactly like editorial. If a product recommendation survives into the draft, check whether it was a read before you publish it under your own name.
Questions people ask
Is this just a transcript with paragraph breaks?
No, and that is the failure mode to watch for. The draft is reordered for a reader — conclusion first, sections by argument rather than by chronology — and it is built to carry the video's specifics, not its phrasing. If the output reads in the video's order, the reorganization did not happen.
Can I publish a post built from someone else's video?
Writing your own piece that credits and links the video is ordinary commentary and it is how most of the web works. Reproducing the transcript at length is not. The draft is built to quote sparingly and link out, and the credit line is generated rather than left to you to remember.
What about videos with no captions?
Ancher still captures frames, the description, the linked sources and the metadata, so a draft is possible but the evidence section will be thin. For a post you intend to publish, prefer a video that has captions or a companion write-up.
How is this different from the free YouTube-to-blog tools?
Most of them read the caption track and nothing else, so anything shown rather than said is gone before generation starts. The second difference is that your video sits in the same workspace as everything else you have saved, so a post can draw on four videos and a report at once rather than on one URL in isolation.
Bring your own YouTube videos
Everything you save lives in one workspace, so the blog post is built from your sources — not from a model's memory of the internet.
Open Ancher →Not a YouTube video? The same blog post also comes from PDF, X thread, LinkedIn post.