Video to text, with full notes
Every line takes you back to the moment it was said.
Your first one is free
Examples
Chapters
1. What is Computation?
This is the first lecture in MIT’s introduction to programming in Python. The most useful takeaway is that computers only carry out the steps you explicitly write down. They don’t infer your goal or automatically recalculate earlier results.
Key points
- The course builds conceptual understanding, programming skills, and problem-solving ability. Beginners should look through the problem sets ahead of time and work through the code shown in class themselves, rather than just watching the demonstrations.
- A computer’s basic jobs are to perform calculations and store the results. Programmers can combine the basic operations built into a language to define new computations.
- An algorithm does more than state a goal. It also gives steps that can be carried out, control flow that determines when to jump or repeat, and a clear stopping condition.
What’s the difference between declarative and imperative knowledge?
Declarative knowledge states a fact, such as the relationship a square root satisfies. Imperative knowledge gives steps for finding the answer. For example, check whether the current guess is close enough. If it isn’t, average the current guess with the original number divided by that guess, then repeat.
Chapters
Do schools kill creativity? | Sir Ken Robinson | TED
In this TED talk, Sir Ken Robinson argues that creativity should be given the same importance as literacy, because children need to be willing to try things and make mistakes if they are to come up with new ways of dealing with an unpredictable future.
Key points
- Education has to prepare children for an unpredictable future. Robinson argues that since children generally have the potential to innovate, creativity should have the same status as literacy.
- Making mistakes is not the same as being creative, but if you are unwilling to risk being wrong, it is hard to come up with original ideas. Treating mistakes as the worst thing that can happen gradually makes children afraid to try.
- Robinson points out that schools usually put math and languages first, followed by the humanities, with the arts last. Within the arts, visual art and music also tend to get more attention than drama and dance. He attributes this hierarchy to industrial-era ideas about employment and universities' emphasis on academic ability.
Why should creativity be as important as literacy?
Schools have to prepare children for the world they will face, but it is hard to predict what the world will look like even a few years from now. Robinson argues that education should not waste the capacity for innovation that children already have.
Chapters
[1hr Talk] Intro to Large Language Models
In this introductory talk, Andrej Karpathy explains how large language models go from predicting the next word to calling tools to solve problems. The key to understanding these systems is distinguishing what the model has learned in its weights, what behaviors fine-tuning has shaped, and what external resources are available at runtime.
Key points
- Running an open-weight model requires the weights and inference code. Llama 2 70B, for example, has 70 billion parameters stored at 2 bytes each, so its weights take up about 140 GB. The talk estimates that training it took about 6,000 GPUs running for 12 days and cost about $2 million.
- Next-word prediction can be seen as a form of lossy compression of the training text: the model learns some facts about the world, but it also imitates document formats and invents plausible-looking details such as ISBNs.
- Pretraining and supervised fine-tuning use the same prediction objective. The main difference is the data: pretraining uses a large amount of internet text, while fine-tuning uses a small set of high-quality question-and-answer examples. Human comparisons of candidate answers can further improve performance.
Why can a model learn facts when it's only trained to predict the next word?
Because some words can only be predicted accurately with background knowledge. The talk uses Ruth Handler's Wikipedia page as an example: to predict the information-rich words on the page, the model needs to learn some facts about her life in its parameters.
Chapters
Sam Altman - How to Succeed with a Startup
Sam Altman's top advice in this startup talk is to build a product so good that users recommend it to their friends on their own. That gets about 80% of the work needed for startup success done.
Key points
- The most important test of a product is whether users recommend it on their own. If you can't explain it in a few sentences, or if few people find it interesting once you've explained it, your thinking may be unclear or demand may not be strong enough.
- Don't just look at how big a market is today; look at whether it can grow quickly. When deciding whether to bet on a new platform, what matters is whether early users use it often and recommend it on their own, not just how many people buy it.
- At least one founder, usually the CEO, needs to be able to get others excited and take charge of recruiting, sales, talking to the media, and fundraising. The vision should naturally become more ambitious as the company grows, rather than being exaggerated from the start.
How can you tell if a product is good enough to spread by word of mouth?
See whether users recommend it to friends without being asked. Whether you can explain the product in a few sentences and whether people find it interesting are also early signs.
Chapters
The Most Misunderstood Concept in Physics
This Veritasium video uses heat engines, sunlight, and the evolution of the universe to explain entropy. The key thing Earth gets from the Sun isn't a net gain in energy, but a steady supply of low-entropy energy that can do more work than the energy Earth radiates away.
Key points
- An ideal Carnot heat engine can run entirely in reverse, but being reversible doesn't mean it's 100% efficient. Its efficiency depends only on the absolute temperatures of the hot and cold sides. High-pressure steam engines in Carnot's day had a theoretical limit of about 32%, but their actual efficiency was about 3%.
- When energy dissipates, the total stays the same, but less of it is available to do work. Clausius summed up the first two laws of thermodynamics this way: The energy of the universe is constant, and the entropy of the universe tends toward a maximum.
- It's not impossible for heat to flow spontaneously from cold to hot. There are simply more microscopic arrangements in which the energy is evenly distributed. In the video's model, increasing the number of atoms in each metal rod from 8 to 80 reduces the probability that the hot rod becomes hotter than it was at the start from 10.5% to 0.05%.
If Earth radiates the energy it absorbs back into space, what exactly does the Sun give us?
The Sun provides a steady supply of low-entropy energy that is more concentrated and can do more work than the energy Earth radiates away. Earth turns a smaller number of high-energy photons into about 20 times as many low-energy photons and emits them back into space.
Full notes
Summary and the big picture
One sentence, key points and a mind map.
Chapter notes
Each chapter written from what was actually said.
What it answers
The questions it answers, with short answers.
Quotes
The lines worth keeping, word for word.
Transcript
Proofread, every line linked to its moment.
Mentioned
People, books, products and ideas it brings up.
Make
Flashcards
Quiz
Slides
Audio overview
Article and outline
Social cards
Sources
YouTube
Videos, Shorts and youtu.be links.
Bilibili
bilibili.com and b23.tv links, one part at a time.
Podcasts
Episode links and direct audio links.
Recordings and files
Record in the browser or upload audio and video.
Featured
YouTube43:051. What is Computation?MIT OpenCourseWareEnglish
YouTube39:481. The Geometry of Linear EquationsMIT OpenCourseWareEnglish
YouTube54:56Justice: What's The Right Thing To Do? Episode 01 "THE MORAL SIDE OF MURDER"Harvard UniversityEnglish
YouTube29:361. IntroductionYaleCoursesEnglish
YouTube57:141. Introduction to Human Behavioral BiologyStanfordEnglish
YouTube20:03Do schools kill creativity? | Sir Ken Robinson | TEDTEDEnglish
YouTube18:34How Great Leaders Inspire Action | Simon Sinek | TEDTEDEnglish
YouTube14:03Inside the Mind of a Master Procrastinator | Tim Urban | TEDTEDEnglish
Hand it to your agent
Ask about it
Answers point to the moments they come from.
Save to your knowledge base
The notes become a document you can edit.
Follow the channel
New episodes come with notes on their own.
Share
A public link to the notes and transcript.
MCP
Other agents can read your transcripts.
Export
Markdown, PDF, SRT and VTT subtitles.
- Which sites and files does it work with?
- YouTube, Bilibili, podcast episodes and direct audio links, plus recordings and audio or video files you upload. If a link can't be read, you're told before anything is charged.
- How long can it be?
- Up to 60 minutes each on the free plan and 600 minutes on a paid plan. If it's longer, you're told before anything starts.
- How long does it take?
- Videos with captions are usually done in seconds; others take about a minute for half an hour of audio. You don't have to wait on the page.
- Who can see the files I upload?
- Only you. The file is deleted when you delete the item, and your content isn't used to train models.
- Can I export or share it?
- Yes. Export the notes and transcript as Markdown or PDF, or subtitles as SRT and VTT, or turn on a public link anyone can open.