Learning from lecture slides with AI – without losing anything

Why summaries barely help

A large review ranked ten common study techniques by their effect. Summarising, highlighting and rereading came out at the bottom, testing yourself and spaced practice at the top (Dunlosky et al., 2013). The bottleneck is not understanding while reading – it is retrieving weeks later.

What that means for AI: don’t have it write texts you only read. Have it build tasks you work through – and flashcards that come back at the right moment.

What a good AI study unit contains

A unit you really learn from has more than explanatory text:

• Explanations with a why: what problem a concept solves, a mental picture, then the exact definition – with an example and a counter-example.
• Exercises that stand on their own: each with everything needed, so it still makes sense later in review.
• Different exercise types: worked calculations, open explain-questions, “find the mistake”, matching – not just multiple choice.
• Flashcards for every definition and theorem.
• Page references per section so you can check the original.

The three typical mistakes – and what helps

1. The AI silently skips pages. With 80 slides in one go, some topics shrink to a sentence. Fix: work in parts (about 40 slides or 15 text pages per unit) and check that every page appears.

2. Formulas and numbers get “smoothed”. Models like to replace unusual notation with a more familiar one. Fix: explicitly ask for the notation of your material and spot-check important parts against the original.

3. The exercises are too easy. A quiz that only tests recognition feels good and achieves little. Fix: ask for open questions, calculations with intermediate steps and exam-level tasks.

How it works with UniApp24

You drop your material – PDF, PowerPoint, Word, photos of the whiteboard, even the subtitles of a lecture recording, several files at once if you like. The app reads everything in your browser and builds a prompt with the text and exact page numbers; that often saves 70–90 % of the tokens compared with uploading a PDF.

You paste the prompt into ChatGPT, Claude or Gemini and paste the answer back. The app checks format, formulas, whether every page is covered and whether every exercise stands on its own. If it finds problems, you get a correction prompt for the same chat. Then it schedules your reviews until the exam.

With fine-tuning you decide what the AI should emphasise: understanding or practice, worked examples, a practice block at the end, links to earlier lectures. More on the methods behind it: study techniques with evidence.

Which model?

The free versions are enough to start. Noticeably better units come from the stronger models – such as Claude Sonnet or Opus, ChatGPT with “thinking”, or Gemini Pro – especially for maths, code and long scripts.

Frequently asked questions

Read on