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 meeting notes
Turn a recorded call on YouTube into meeting notes
Paste a link. Decisions, owners and open loops — with anything unattributed left honestly unassigned.
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.
ContextDecisionsAction itemsOpen questions
Meeting notes
A record of what was decided, what is owned, and what is still open.
Ancher reads a recording for commitment language rather than for topic changes, so the output is a list of decisions and actions instead of a transcript with headings. Because a caption track carries no speaker labels, an action nobody verbally claimed is written as unassigned rather than attributed to a guess, which is the difference between notes you can circulate and notes that create commitments nobody made.
The honest version
There are no speaker labels, so we do not pretend there are
Caption tracks and machine transcription both return words without saying who spoke them. Tools that assign owners anyway are inferring from context, and a wrong owner on an action item is worse than a blank one — it looks authoritative, it gets circulated, and the person named never sees it. Ancher only assigns an owner when someone verbally claims the task.
One second of one video, at 41:18, read two ways.Illustrative. The figure is invented; the gap it falls into is not.
...so we hold the launch until the numbers recover — I'll own that, and we revisit at .
On the shared screen — not readThe specific never reaches the draft
Meeting notes
On the shared screen
Review 14 Nov
from 41:18
Lands in the draft, with its timecode
- Owners come from explicit acceptance in the words, so an assignment is traceable to a sentence someone actually said.
- Anything nobody claimed is written UNASSIGNED, which turns a silent gap into a visible one.
- Shared screens are sampled as frames, so the figure that drove a decision stays attached to it.
What comes out
Meeting notes, section by section
A record of what was decided, what is owned, and what is still open.
Context
What the call was for, drawn from the opening minutes, which is where the purpose is stated if it is stated at all.
Decisions
Only positions someone settled, quoted with a timecode so the wording can be disputed precisely rather than in general.
Action items
Tasks with an owner where one was verbally claimed, and UNASSIGNED where none was — never a guess.
Open questions
Threads raised and never closed, which is usually the reason the next meeting exists.
Where each line came from
The screen or figure that informed each decision, so a reader can see what the room was looking at.
Recording-to-meeting-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.
From this recording, write meeting notes. (1) Record a decision only where someone stated a settled position; quote the sentence with its timecode. (2) Create an action item only where a person verbally accepted it; otherwise write the owner as UNASSIGNED. (3) Never infer who was speaking — the transcript has no speaker labels. (4) List every topic raised and not resolved under Open questions. (5) Attach any on-screen figure to the decision it informed. Do not create commitments from politeness such as "sounds good".
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 | Action items | Commitment phrases — "I'll take that", "by Friday" — are what separate a task list from a topic list, and they are all in the words. |
| Caption track with timings | Decisions | A decision is recorded only where someone stated a settled position, not where a topic simply got the most airtime. |
| Key frames from the video | Where each line came from | A shared screen is the number the room was actually reacting to, and it exists nowhere in the audio. |
| Caption track with timings | Open questions | Segment timings expose threads that were raised and never returned to, which no summary pass surfaces on its own. |
| Title, channel, duration, date | Context | Title, channel and date give the notes provenance, which matters more for a public recording than an internal one. |
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 meeting notes. Ancher fills context, decisions, action items, open questions, where each line came from 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
- A published recording is edited. The moment a decision was reversed can be cut, so notes from a public video should never be treated as a system of record for your own team.
- Without speaker labels, an action item's owner comes only from someone saying so. Verify every owner before circulating, and expect several to be legitimately blank.
- Politeness reads as agreement in text. "Sounds good" is not a decision, and notes that treat it as one manufacture commitments that propagate into the next meeting.
Questions people ask
Why can't it tell who said what?
Because the transcript carries no speaker information — Ancher stores what was said and when, not who said it. Inferring owners from context would produce confident, occasionally wrong attributions, which is the more damaging error for a document people act on.
What about my own team's calls?
Same workflow, better inputs. Upload the recording file rather than publishing it, and you keep the timings without putting an internal conversation on a public platform. The speaker limitation is the same either way.
Can it handle a recurring series?
Yes, and the value compounds. Across a series, an action item that reappears unresolved in three consecutive meetings is a pattern no single set of notes can show you.
Bring your own YouTube videos
Everything you save lives in one workspace, so the meeting notes is built from your sources — not from a model's memory of the internet.
Open Ancher →Not a YouTube video? The same meeting notes also comes from recording, web page.