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 notes
Turn a YouTube video into notes
Paste a link. You get points you can search six months from now, each one still attached to its timecode.
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.
One-line summaryKey pointsTerms and definitionsQuotes worth keeping
Notes
Structured notes you can search and revise from later.
Notes are for finding things again, so Ancher writes them as discrete points rather than as a flowing summary, and keeps the timecode on each one. Terms the speaker defines get pulled out separately from claims they make, because those two things get looked up for different reasons, and anything the video raises without settling is filed as a follow-up rather than quietly dropped.
The honest version
A summary is a dead end; notes are an index
Most video tools hand back a paragraph. A paragraph is fine to read once and impossible to search, which is why the summary you generated in March is no help in September. Notes are worth making only if each line stands alone, carries where it came from, and sits next to everything else you have saved on the subject.
One second of one video, at 22:41, read two ways.Illustrative. The figure is invented; the gap it falls into is not.
...so the retrieval step matters more than the model here, and we saw that clearly once we on the eval set.
Recall @10, after reranking — not readThe specific never reaches the draft
Notes
Recall @10, after reranking
0.91
from 22:41
Lands in the draft, with its timecode
- Each point keeps its timecode, so a note found months later can be re-checked in one click.
- Definitions are separated from claims, because you look up a term and an assertion for different reasons.
- Notes land in the same workspace as your other sources, so a search crosses this video and everything near it.
What comes out
Notes, section by section
Structured notes you can search and revise from later.
One-line summary
One line stating what the video argues, written so that reading it alone tells you whether to open the rest.
Key points
Discrete claims, one per line, each with a timecode. Not a paragraph — a paragraph cannot be scanned or reordered later.
Terms and definitions
Any term the speaker defines, captured at the moment of definition rather than from a recap where it has been compressed.
Quotes worth keeping
Lines worth keeping verbatim, kept verbatim, because a paraphrased quote is useless the moment you need to cite it.
Follow-ups
Questions the video raises and does not answer, which is the part a summary always deletes and the part you actually come back for.
Video-to-notes 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.
Turn this video into notes. (1) One summary line that tells a reader whether to open the rest. (2) Discrete points, one claim per line, each with a timecode; never write a paragraph. (3) A separate list of terms the speaker defines, taken from where the term first appears. (4) Verbatim quotes only where the exact wording matters. (5) A follow-up list of questions the video raises and does not answer. Every point must stand alone if moved elsewhere.
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 | Key points | Caption segments already break at natural pauses, which is a better unit for a note than a sentence a model invents. |
| Caption track with timings | Terms and definitions | A definition is caught where the term first appears, not from the closing recap where the speaker has already shortened it. |
| Key frames from the video | Key points | A figure shown on a slide becomes its own note; otherwise the only record of it is a frame nobody will scrub back to. |
| Title, channel, duration, date | One-line summary | Channel and date sit on the summary line so a note found later carries its own provenance without a lookup. |
| Description and linked sources | Follow-ups | Unanswered questions are paired with the creator's own links, which is usually where the answer was going to come from. |
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 notes. Ancher fills one-line summary, key points, terms and definitions, quotes worth keeping, follow-ups 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
- Notes that read as prose have failed at the only thing notes are for. If a line cannot be moved elsewhere and still make sense, it is a sentence, not a note.
- Auto-captions garble technical terms badly, and a wrong term in a definition is worse than a missing one. Check any term you plan to rely on against the frame.
- A one-hour video yields far more notes than you want. Decide the question before generating; notes taken without one are just a longer transcript.
Questions people ask
How is this different from a summary?
A summary compresses the video into something readable once. Notes decompose it into pieces that survive being searched, reordered and quoted later. If you will read it once and move on, take the summary — notes are for material you expect to come back to.
Can I search across many videos at once?
Yes, and that is where the format pays off. Notes from every video you have saved sit in one workspace, so a search for a term surfaces every place it was defined or disputed rather than one video at a time.
Does it keep the speaker's wording?
Only in the quotes section, and there it is kept verbatim. Everywhere else the points are rewritten to stand alone, because a note that depends on the surrounding sentence is not usable on its own.
Bring your own YouTube videos
Everything you save lives in one workspace, so the notes is built from your sources — not from a model's memory of the internet.
Open Ancher →Not a YouTube video? The same notes also comes from PDF, X thread, LinkedIn post.