Cutlist

Only the minutes that matter.

YouTube has the explanation you need. It is buried in minute 34 of an hour-long video you have not found yet. Say what you want to understand and how long you have: Cutlist works out what the topic is made of, fits as much of it as your time allows, finds the clearest explanation of each piece, and orders them so each one is ready for the next.

What do you want to understand?

Covers AI and everything around it: machine learning and deep learning, LLMs and agents, image, video and voice generation, and the evals, fine-tuning and product skills that go with them.

Say what you want to understand first.

Demo mode. Stages are timed like a real run, but the plan is one captured earlier and no model is being called.

Nothing in mind? Take one of these.

Each one compiles a full plan

You bring a question
and a deadline.

A search box gives you ten hour-long videos and wishes you luck. But the harder problem comes first: knowing what the topic is even made of, and how much of it you can realistically learn before the time runs out. Cutlist answers that first and then fills it in. The unit of YouTube is the video, the unit of learning is the idea, and they do not line up, so you pay a search-and-skip tax on every video and pay it again on the next one.

  1. 01

    Ask in plain language

    A topic, a question, or the thing you are stuck on. "I know nothing about AI, where do I start" works as well as a keyword does.

  2. 02

    Say how long you have

    Thirty minutes before a meeting, or a free Sunday. Your time is the constraint the plan is built around, so it fills the hour without running past it.

  3. 03

    The topic comes apart

    Cutlist works out the ideas the topic is actually made of, then how many of them fit the time you gave it. You see that skeleton before you see a single video.

  4. 04

    Get a plan, not a results page

    Every entry opens the creator's own video at the second it starts being useful. Nothing repeats, and nothing is filler.

What a plan does that a search box cannot

  • You watch minutes, not videos. A search result is a 52-minute video with the right words in the title. Finding the four minutes that actually answer you is still your job, and you do it again on the next video. Here that work is already done: every entry opens at the second the answer starts.
  • You never sit through the same idea twice. Ten creators explain embeddings, and nine of them explain it much the same way. Watching all ten is not ten times the understanding, it is nine repeats. Overlap is caught by meaning rather than by title, so each idea earns its place once and the rest of your hour goes somewhere new.
  • You are told what is missing. When part of your topic has nothing good enough behind it, that part is named on the page and struck through. Your time is never quietly filled with a loosely related video, so you finish knowing exactly what you still have to go and find elsewhere.

How it is built

Eight stages. Three call a model. Five are plain code.

    Most of this is not a model. Deduplication is a similarity problem, so embeddings do it. Filling a time budget is a constraint problem, so an algorithm does it. Handing both to a model was the first thing tried, and it returned a plan with the same explanation in it seven times.

    How much time have you got?

    One domain so far: AI, from machine learning basics to LLMs, agents and generative media. Ask it something in that world and it will tell you honestly what it can and cannot cover.