YouTube Summary just passed 100,000 summarized videos. That number is hard for me to picture. Three years ago, this was a side project that took a YouTube link and tried to turn its subtitles into something readable. It is still a side project, but there are now 100,000 videos in the collection.
The first one is still there
You can read the very first summary. It is a short paragraph about a Modern C++ talk, followed by a speaker's name. There is no category. The formatting is nothing like what you see on newer pages.
I like that it is still there. It makes the progress visible. We started with a much shorter, rougher answer and kept improving it one step at a time.
The site looked different too. Here is a screenshot of the home page from April 30, 2024. Click it to explore that capture in the Wayback Machine.
Back then, the project used Microsoft's Semantic Kernel. In AI time, that sounds ancient. Microsoft has since introduced Agent Framework as its successor, although Semantic Kernel still exists under that name. It has only been about three years, but it somehow feels like yesterday and a different era at once.
From one paragraph to thirteen hours
We added categories, improved the general prompts, and then wrote prompts tailored to different kinds of videos. A cooking tutorial needs a different summary from a lecture or a product review. Long videos also needed a way to be processed in parts and brought back together into a useful summary.
One example is this 13-hour AWS certification course summary. It is a long, structured page covering the course's main ideas and practical steps. That single video cost around $0.13 in AI processing. Compare it with the first summary and you can see both how far the output has come and how much more the tool can handle.
The money, honestly
When we crossed 30,000 summaries, ads were not quite covering the AI bill. They now do. AI summarization costs roughly $100 a month, and ads bring in roughly $150 a month. Those are approximate monthly figures, not a guarantee that every month looks the same.
That means the AI cost is covered. It does not mean the ads fully pay for the server too. I use that server for other things, so I do not count its entire cost as a YouTube Summary expense. I am happy with where that leaves the project: free to use, with its main running cost paid for.
What comes next
It is fun, and it has its challenges. I want to keep it running smoothly without letting a side project consume all my time. For now, it is doing just that, and I plan to keep it going.
When I have time, I would like to work on features people have asked for, including summary translation. I am not promising a date. In the meantime, you can still summarize a video, browse what others have found, or click back to that first little C++ summary and see where this all started.
