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 slide deck
Turn a YouTube video into a slide deck
A talk already has an argument. Paste it and get the deck that argument should have had.
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.
Narrative arcSlide-by-slideData slidesAppendix
Slide deck
A paced presentation: one argument per slide, built to be talked over.
A conference talk and a slide deck are different objects: a talk can hold a digression for four minutes, a deck cannot hold one for a single slide. Ancher finds where the speaker's argument actually turns, gives each turn one slide, lifts the charts that were already on screen as data slides rather than describing them in bullets, and writes speaker notes from what was said around each one.
The honest version
A talk's structure is not a deck's structure
Transcribing a talk into slides gives you forty slides of bullets, because speech is redundant by design — a speaker repeats to help a listener who cannot scroll back. A deck has the opposite constraint. The work is deciding which of the speaker's twelve points were the argument and which were scaffolding, and that decision is made from where they spent time, not from what they said first.
One second of one video, at 24:18, read two ways.Illustrative. The figure is invented; the gap it falls into is not.
...and if you look at this breakdown here, the second cohort is doing something completely different.
Chart on screen — not readThe specific never reaches the draft
Slide deck
Chart on screen
Cohort retention, W1–W12
from 24:18
Lands in the draft, with its timecode
- One slide per turn in the argument, so a forty-minute talk becomes ten or twelve slides rather than one per topic sentence.
- Charts shown on screen come across as images with their timecodes, instead of being flattened into a bullet describing a chart.
- Speaker notes hold the sentences the slide is standing on, so whoever presents it can defend the claim without watching the source.
What comes out
Slide deck, section by section
A paced presentation: one argument per slide, built to be talked over.
Narrative arc
The through-line in one line, taken from the claim the talk keeps returning to rather than its opening promise.
Slide-by-slide
One slide per turn in the argument, each with a single assertion as its title.
Data slides
The charts and tables that were already on screen, kept as images with the timecode they came from.
Appendix
The digressions worth keeping but not worth a slide, so they are available if asked but not in the flow.
Speaker notes
What the speaker actually said around each slide, so the deck can be presented by someone who was not there.
Talk-to-slide-deck 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 talk into a slide deck. (1) State the through-line in one sentence, taken from the claim the speaker returns to rather than the one they open with. (2) Give one slide per turn in the argument, titling each with the assertion it makes, not with a topic. (3) List every chart or table shown on screen as a data slide with its timecode. (4) Put digressions in an appendix rather than the main flow. (5) Write speaker notes from what was said around each slide. Keep the total under fifteen slides and say what you cut.
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 | Narrative arc | Where a speaker returns to the same claim is visible in the caption timings, and it is the through-line whatever the title said. |
| Key frames from the video | Data slides | The chart is the slide. Describing it in a bullet is the single biggest thing lost when a video becomes text. |
| Caption track with timings | Speaker notes | Speaker notes are the transcript around a slide, trimmed of the repetition that speech needs and reading does not. |
| Caption track with timings | Slide-by-slide | Pauses and subject changes in the timings mark where the argument turns, which is where a slide break belongs. |
| Title, channel, duration, date | Appendix | Chapter markers, where the creator set them, separate the main line from the asides more reliably than any heuristic. |
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 slide deck. Ancher fills narrative arc, slide-by-slide, data slides, appendix, speaker notes 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 deck with one slide per topic sentence is a transcript in a 16:9 box. If the slide count is close to the minute count, the argument was never extracted.
- Charts pulled from frames carry the source's own styling and axis labels. Re-cut them before presenting anything externally, and keep the timecode so the original is findable.
- Presenting someone else's argument as your own deck is a credit problem, not a formatting one. The source credit belongs on the deck, not only in your notes.
Questions people ask
How many slides should a talk become?
Roughly one per turn in the argument, which for a forty-minute conference talk is usually ten to fourteen. If you are getting thirty, the extraction is following the transcript rather than the reasoning.
Does this produce a PowerPoint file?
It produces the deck's content and structure — titles, the assertion each slide makes, the data slides and the speaker notes. Getting that into your own template is a paste, and it is the part worth doing by hand anyway.
What happens to the demo section of a talk?
Demos become data slides built from frames, because that is where the demo actually lives. The caption track for a demo is mostly the speaker narrating clicks, which is worth very little on a slide.
Bring your own YouTube videos
Everything you save lives in one workspace, so the slide deck is built from your sources — not from a model's memory of the internet.
Open Ancher →Not a YouTube video? The same slide deck also comes from X thread, LinkedIn post, Instagram post.