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 financial analysis
Turn a YouTube video into a financial analysis
Every figure, tied to the sentence or the slide it came from — and flagged when it came from neither.
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.
Headline figuresEvidence per figureTrendRisks
Financial analysis
Figures checked against the sentence or table that supports them.
Numbers in a video arrive two ways: spoken, where auto-captions mangle them, and shown on a slide, where a caption-only tool never sees them at all. Ancher collects both, attaches the timecode and the form each figure appeared in, and separates figures the speaker sourced from figures the speaker asserted — because in a video those look identical and only one of them is usable.
The honest version
A video is commentary on numbers, not a source of them
Presentations and briefings restate figures that live in a filing, a dashboard or a deck. By the time a number reaches a video it has been rounded, simplified for an audience, and sometimes recalled from memory. Treating that number as the number is how errors travel. Every figure carries where it appeared and in what form, so the ones worth checking against a primary are obvious.
One second of one video, at 09:57, read two ways.Illustrative. The figure is invented; the gap it falls into is not.
...gross margin came in at about which is where we said it would land.
On the slide — not readThe specific never reaches the draft
Financial analysis
On the slide
Gross margin 62.4%
from 09:57
Lands in the draft, with its timecode
- Spoken and on-screen figures are labelled separately, because auto-captions get digits wrong far more often than words.
- A figure the speaker attributed is kept apart from a figure the speaker simply stated, which is the difference between a citation and a recollection.
- Nothing is annualised, converted or extrapolated. Derived numbers are the reader's to make, with the source in view.
What comes out
Financial analysis, section by section
Figures checked against the sentence or table that supports them.
Headline figures
The figures the speaker leads with, each marked spoken or on-screen so their reliability is visible at a glance.
Evidence per figure
One line per figure: the exact words or the frame it appeared in, plus its timecode.
Trend
Only comparisons the speaker actually made, never a trend assembled from figures they never put side by side.
Risks
Caveats and hedges as spoken, including the ones delivered quickly at the end of a sentence.
Sources
Whatever the description links to, which is usually where the real numbers are.
Video-to-financial-analysis 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.
Extract a financial analysis from this video. (1) List the headline figures, marking each SPOKEN or ON-SCREEN. (2) For every figure give the sentence or the frame it came from and its timecode. (3) Note only comparisons the speaker made themselves; do not assemble trends they did not state. (4) Record caveats and hedges in the speaker's words. (5) List the description links as sources. Do not annualise, convert, extrapolate or calculate anything, and mark every figure the speaker did not attribute.
The detail, if you want it
How each part of a YouTube video becomes a section
| From the YouTube video | Into | Why |
|---|---|---|
| Key frames from the video | Headline figures | The precise figure is nearly always on a slide while the spoken version is rounded, so the frame is the more accurate of the two. |
| Caption track with timings | Evidence per figure | The sentence around a figure is what says whether it is sourced or asserted, and that context disappears in any summary. |
| Caption track with timings | Risks | Hedges are spoken fast and rarely make it onto a slide, so the transcript is the only place they survive. |
| Description and linked sources | Sources | A description linking to the filing or deck turns a commentary into something checkable against a primary. |
| Title, channel, duration, date | Trend | The publish date fixes the period being discussed, without which a quarterly figure is just a number. |
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 financial analysis. Ancher fills headline figures, evidence per figure, trend, risks, sources 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-captions misread digits far more often than letters. Any figure that only ever appeared as speech should be checked against the frame or the filing before it is quoted.
- This is an extraction of what was said, not a valuation. No multiple, growth rate or projection is calculated from these figures, and adding one on top of video-sourced numbers compounds the error.
- A figure with no source in the video and no link in the description is an assertion. It stays in the analysis marked as one, and it does not belong in a model.
Questions people ask
Can I build a model from this?
You can build the list of things to verify, which is the useful half. The figures come with their form and timecode so a primary source check takes minutes rather than a rewatch, and that is the step this is meant to save.
What if the video shows a chart but never says the number?
That is the common case, and it is exactly what frame reading is for. The value comes across from the chart with its timecode, marked on-screen, so you can see it was never spoken aloud.
Is an earnings call video as good as the transcript?
No. The official transcript is the better source for what was said, and the filing is the better source for the numbers. A video adds what was on the slides and how the hedges were delivered, which neither of those contains.
Bring your own YouTube videos
Everything you save lives in one workspace, so the financial analysis is built from your sources — not from a model's memory of the internet.
Open Ancher →Not a YouTube video? The same financial analysis also comes from X thread, PDF, recording.