How to chat with your notes using AI in Filmina (and see where every answer comes from)

If you're looking up how to chat with your notes using AI, here's the short answer: give your own material (the lecture slides, the class PDF, your notes) to a tool that answers only from that, ask it specific questions, and check where each answer comes from. What matters isn't whether the answer sounds good. It's whether it's the answer from your class, with the definition and the example your professor actually used.

That's a big difference from asking a general chatbot. A general chatbot answers with what it picked up from the whole internet, which sometimes matches your course and sometimes doesn't; your material has the version that'll be on the exam. A good answer tells you which slide it came from, so you can go look at it.

In this post I'll go through what to ask any AI tool, how I do it in Filmina step by step, and which questions actually help, using an example lecture from Intro to Microeconomics: price elasticity of demand, a deck of about thirty slides with definitions, formulas and several graphs.

Why ask your material instead of the internet

When you study alone, the questions show up when there's nobody to ask: late at night, on a Sunday, with the midterm on Tuesday. Searching online works for some of them, but it has three problems when you're studying from a specific lecture:

1. It gives you a different version. Elasticity means the same thing everywhere, but how you calculate it (point, arc, midpoint) and which example you use change from one textbook to the next. The exam wants your professor's.
2. It can't see your slide. If what's confusing you is a graph with two demand curves, you have to describe it in words, and half of it gets lost.
3. It doesn't tell you what's yours and what isn't. An answer blends what was in your lecture with whatever it added, and you have no way to know which part will be on the exam.

Asking your own material fixes all three: it answers from what you were given, it can look at the slide, and it shows you where the answer comes from.

How to chat with your notes using any AI tool

If you use a general AI chatbot, you can do this well with a few precautions:

• Upload the lecture file. Don't ask it to go find it.
• Set the rule up front. "Answer only from this PDF. If something isn't in it, tell me instead of filling it in."
• Ask for the source. "End each answer with the slide number it comes from."
• Ask one thing at a time. "What does it mean for demand to be elastic?" works better than "explain the lecture."
• Check an answer against the PDF. Especially the one that surprises you most.

The catch is that every conversation is an island: you have to upload the file again, the chat gets buried among others, and when your question is about a graph, you still have to describe it. I went through how to get more out of a general chatbot in how to use ChatGPT to study.

How to chat with your notes in Filmina, step by step

Filmina, a web app that explains your lecture slides out loud, has a chat for asking questions about your own material. Each course is a space with its files (slide decks, PDFs, books, audio, video and your notes), and Filmina answers from the files in the space you have open.

1. Open the chat. Top right, next to the timer, there's Ask Mina, on every screen. The shortcut is ⌘J (Ctrl+J on Windows). It opens on the side, over whatever you were looking at; on your phone it lives in the ⋯ menu at the top.
2. Check what it's looking at. Above the input box, a small line says what it'll answer from: "Slide 12," "Page 4." With an empty chat, up to three suggestions show up depending on the screen; on a slide, for example, "Explain" or "Give me an example" with that slide's title.
3. Ask. Type it, or dictate it with the mic next to the input box. Dictated text lands in the box without sending, so you can fix a word it misheard.
4. See where it comes from. Above the answer it says From your files, From your notes or From your material. If nothing of yours covers it, it still answers and flags it: Beyond your material. Below the answer are the slides it used, as thumbnails; tap one to see it full size, and Study it takes you to that slide in the file.

You don't have to pick where it looks. From inside a file, it searches that file first; if the answer isn't there but is in another lecture from the course, it moves on to the rest and tells you which file it came from. If the question is about the whole course ("how is this lecture different from the one on supply?"), it searches the whole space from the start.

It also remembers each conversation by place: every file, summary or note has its own, so "and an example?" makes sense after your last question, and what you discussed about another file doesn't bleed in.

Asking right on the slide you don't get

What I use most is asking in the moment, while Filmina is explaining the lecture. On the slide comparing elastic and inelastic demand, the bubble in the dock (Ask about this slide) or the ? key opens the chat and pauses the voice. The box reads "What didn't you get?" If I had highlighted the bit of the explanation that lost me, it comes along as a quote above my question, so I don't have to type it again. When I close the chat, the explanation picks up from the start of that paragraph.

Sometimes a question is easier to point at than to type. That's what Point is for, to the right of the line above the input box: you circle the part you don't understand with your mouse, finger or stylus (the flatter curve, the midpoint formula) and ask about that. Filmina gets the cropped area as an image along with the whole slide, and answers from that file.

In a long PDF or a book it works the same way with text: select it and tap Ask in the toolbar that pops up. The passage goes along as a quote, with where it came from. All the details are in Ask Mina.

What to ask so it actually helps on the exam

Having someone to ask isn't enough: some questions teach you something and some leave you exactly where you were. Asking yourself why something is the way it is, instead of just what it is, is a technique that a major review of study techniques rated as moderately useful, ahead of rereading or highlighting. With the elasticity lecture, these are the kinds of questions that work best for me:

• Why. "Why is demand for a good with no substitutes inelastic?"
• The difference. "How is elasticity different from the slope of the curve?"
• Another example. "Give me a different example from the one on slide 9."
• What's likely on the exam. "Which problem from this lecture is most likely to show up on the midterm?"
• Flip it around. "Quiz me on this slide," and you answer without looking.

In Filmina that shows up later. What you asked while it was explaining is among the first things to come up in Practice, and in Review the concepts from those slides come back first, marked "You asked about this." And if you ask for something the app can do, it answers with a card you can act on in one tap: a practice set on what you missed, a jump to the slide it's talking about, or saving the answer to your notes. How those questions get built is in how to make practice questions from a PDF, and how it all fits with the rest of the app is in how Filmina works.

Its limits, up front

• It needs an internet connection and an account. It runs in your browser and the answers are generated in the cloud.
• It doesn't search the web. It answers from your files, your notes and what the model knows about the subject. It may not know about things that happened after the model was trained, and anything that doesn't come from your material gets flagged as Beyond your material.
• Questions have a monthly cap. It depends on your plan, and the free one gets fewer; details are on the pricing section. Each question can be up to 500 characters: for something long, select it or point at it instead, since that goes along separately.
• Scanned slides read worse. If the deck is a photo of each slide, Filmina recognizes the text from the image, and that can trip over small print or formulas. If you have the original file, upload that.
• It only knows what's in your material. If your professor explained something in class that isn't in the PDF, it doesn't have it, unless it's in your notes.

Frequently asked questions

Can I ask an AI questions about my own notes?

Yes. Tools that work with your files answer from what your slides, PDFs and notes say. Look for one that tells you which file and page each answer comes from, so you can check it.

How do I stop AI from making things up about my PDF?

Tell it to answer only from the file and to say when something isn't there. Filmina flags anything that doesn't come from your material as Beyond your material and shows the slides it used for each answer.

Can I ask about a graph on a slide?

Yes, if the tool reads the slide as an image. In Filmina, you use Point to circle the graph or formula and ask about that part; it gets the cropped area along with the whole slide.

Does asking AI questions replace studying?

No. It's great for getting unstuck in the moment, but what makes things stick is trying to recall them and practicing. That's why it helps to turn the answers into questions and review whatever gave you trouble.