Bilibili videos · Summaries · Source timestamps

Bilibili video summaries with chapter timestamps

Read the main ideas by chapter, then follow timestamps back to the video to check them.

We check existing captions first. A new summary requires sign-in and AI allowance, and starts only after confirmation in the reader. Speech transcription is not started automatically.

Example: Compression is Intelligence

This example comes from Compression is Intelligence, Part 1: Reinventing Entropy, uploaded to Bilibili by 3Blue1Brown. The video lasts 32 minutes and 20 seconds.

The original example uses a saved English speech transcription with 217 captions, an AI summary and its Chinese translation. These are not Bilibili platform captions or a creator-written summary. The saved summary and Chinese translation were generated on September 18, 2026; the source and excerpts were checked on October 20, 2026. That saved edition has 16 chapters; this page presents an English rendering of five selected points. The full reader may offer a newer saved summary.

Reading guide: The explanation starts with robot-instruction codes and develops information content, language prediction and entropy. Read the points, then revisit the derivation at the chapter timestamps. These are summary notes, not verbatim quotations.

00:00 Compression limits and information theory

00:59 Information theory establishes a mathematical equivalence between prediction and compression.

04:29 Prefix codes for robot instructions

05:47 The weighted sum gives an average of 1.75 bits per instruction, improving on a fixed 2-bit code.

15:39 Information as negative log probability

16:14 Shannon defines an event's information content as the negative logarithm of its probability, p. The definition remains useful when perfect compression is impossible and p is not a power of 2.

21:14 Shannon's n-gram limits and Betty's test

22:34 When Betty guesses incorrectly, Shannon records the correct letter; when she guesses correctly, he records a dash.

25:09 Entropy as average information per symbol

25:19 Entropy measures the average information content per symbol.

Chapter links locate the beginning of a topic; point links locate the relevant explanation. Bilibili's player may handle seeking differently across clients. You can also search the transcript to find the exact words.

The reader can display saved summaries and translations in the selected language. This example reuses saved results; it does not mean every Bilibili video already has a summary, or that generating again would produce identical wording.

How can I check the “1.75 bits on average” point?

The point links to 05:47, when the weighted-sum explanation starts. The saved transcript reaches the number 1.75 at 06:00. A timestamp locates the passage, not necessarily the precise second when a number is spoken.

  1. Search the original transcript for 1.75 and read the calculation around the match.
  2. Review the four instruction probabilities at 03:51 and the independent-sampling assumption at 04:00.
  3. Keep the conditions: probabilities are 1/2, 1/4, 1/8 and 1/8, with code lengths of 1, 2, 3 and 3 bits. The average is the weighted sum of those probabilities and lengths; it is not a universal figure for any text.

This verification guide is Vidleaf editorial text, separate from the AI summary. When quoting, distinguish the speaker's words, automatic transcription and your own explanation.

From a Bilibili link to a verifiable summary

1

Confirm the video and captions

Paste a BV ID, full video URL or b23.tv share link. Keep the p parameter for multipart videos and check the episode, title and caption source.

2

Generate only when needed

Reuse saved results when available. To create a new summary, sign in and review AI allowance before confirming. Choose the desired summary language or a saved translation in the reader.

3

Verify and organize

Browse topics by chapter, search for specific statements and check the original video. Export summary notes as Markdown. For direct quotations, select complete sentences from the transcript and keep the source.

Summary, captions or transcription?

You needSuggested action
An overview of a long videoRead the captions, open or generate a chapter summary, then verify the points that matter.
Exact words or a TXT/SRT downloadUse the Bilibili caption reader; no summary is needed first.
Text for a video confirmed to have no captionsReview audio processing, duration and transcription minutes separately. Confirm transcription, then decide whether to summarize.
A rate limit, verification request or failed readCheck the link and access, then retry later. A failed read does not prove captions are missing and must not automatically trigger paid transcription.

Common questions

Can I read or generate summaries without signing in?

You can read this example directly. Available saved summaries can be reused. Generating a new summary requires sign-in and AI allowance. Submitting a link here starts the caption-reading flow; it does not immediately call a model.

Yes. Read, search and export subtitled videos without signing in, with a daily limit on new reads. Signed-in users can read subtitled videos without a limit, use transcription, summaries and translation within their allowances, and save history and personal labels. Email code or Google; no phone number needed. See allowances and plans.

What about videos without captions or with multiple parts?

A summary needs readable text. Retain the p parameter and check the episode in the reader. This example covers only the linked episode. After captions are confirmed absent, you may separately confirm speech transcription and then summarize its result. Pasting a link does not remove access restrictions.

Are summaries, translations and reading history public?

Video captions, summaries and translations are cached per video and can be reused by other readers, including guests when saved results are available. They are not private personal notes. Personal tasks, history and speaker labels are not published as public content.

This page is an operator-selected product example. Generating a summary does not automatically create a search-indexed page. See the privacy policy.

Continue reading