5 AI Tools I Regret Not Using Earlier as a Student
By My Code Diary
I spent three years in university doing things the hard way. I’d re-read the same dense textbook chapter four times and still not understand it. I’d spend two hours formatting a lab report when I should have been analyzing data. I’d stare at a blank screen trying to figure out how to start a research paper, kill an hour on Reddit, and convince myself that was “thinking time.”
Then I graduated, started working with AI tools seriously, and had that specific kind of regret, the “where was this two years ago” kind. The tools I’m about to share aren’t ChatGPT. They’re not the obvious ones. These are the ones that would have genuinely changed how I studied, not just how fast I typed.
Here’s what I wish someone had handed me on Day 1.
1. NotebookLM, The Tool That Reads Your Sources So You Don’t Have To (Poorly)
The number one mistake students make when researching is reading everything linearly. You download 12 PDFs, open them all, and read each one from abstract to conclusion hoping meaning accumulates like sediment.
It doesn’t. You just get tired.
NotebookLM (by Google) lets you upload your actual source documents PDFs, Google Docs, URLs and then have a conversation with them. Ask it to find contradictions between two papers. Ask it to summarize the methodology of one study. Ask it to pull every paragraph that mentions a specific variable.
What makes it genuinely different from “paste into ChatGPT” is that it only answers from your uploaded sources, and it cites them. You get a response with footnotes pointing back to the exact passage. For academic work, that’s not a nice-to-have, it’s the whole game.
Pro tip: Upload your lecture slides alongside the textbook chapter. Ask NotebookLM to explain how the professor’s examples connect to the theory. You’ll understand both better in ten minutes than you would in two hours of re-reading.
If I had used this in my research methods course, I would have written a better literature review in half the time. Instead, I used highlighters. Many, many highlighters.
2. Perplexity, For When You Need Answers, Not a Search Results Page
Here’s a problem every student knows: you search for something, you get a list of links, you click three of them, two don’t load properly, one is from 2009, and you’ve now lost twelve minutes and learned nothing.
Perplexity fixes this by combining a search engine with an LLM. You ask a question, it searches the current web, and gives you a synthesized answer with numbered citations you can actually verify. No ads. No SEO-farmed blog posts about nothing.
The part I use constantly is the Focus feature. You can restrict searches to academic papers (via sources like PubMed or Semantic Scholar), to Reddit threads, to YouTube, or to general web. For a student, the academic mode alone is worth it. Ask a question about a statistical method, and instead of getting a tutorial blog, you get references to actual published research.
It also handles follow-up questions in context, which means you can go three or four levels deep into a topic without starting over.
The mistake students make here is using it like a search engine single queries, shallow. Use it like an expert you’re interviewing. Ask the obvious question, then ask the follow-up, then ask “what are the main criticisms of this?” That’s where it earns its place.
3. Anki + AI-Generated Cards, Spaced Repetition Without the Setup Tax
If you’ve heard of Anki, you probably also know the dirty secret: making good flashcards takes almost as long as studying the material. The result is that most students either don’t use it, or they use pre-made decks that don’t match what their professor actually tested.
The fix is simple and I’m annoyed I didn’t do this earlier. You generate the cards with an AI.
Give the model your notes or a textbook section, and ask it to produce Anki-style flashcards in a specific format. You can even get it to write cards in a way that tests application, not just recall:
Prompt: "Here are my notes on ANOVA. Generate 15 Anki flashcards in Q&A format.
For each card, ask a question that tests understanding, not just definition.
Include at least 5 cards that present a scenario and ask which statistical
test to use and why."
Then paste the output into Anki (or use a CSV import). The whole process takes ten minutes. You’re not outsourcing the learning you’re outsourcing the card formatting, which was never the valuable part anyway.
The research on spaced repetition is solid. Students who use it consistently outperform those who don’t on long-term retention. The bottleneck was always setup time. That bottleneck is gone.
4. Elicit A Research Assistant That Screens Papers for You
If you’ve done any kind of literature review, you know the pain: you need to find papers that are relevant, recent, methodologically sound, and actually answer your question. The process usually involves searching Google Scholar, clicking through abstracts, downloading papers, realizing half of them don’t apply, and repeating.
Elicit automates the boring part of that process. It’s an AI research assistant that searches across a database of academic papers, extracts specific columns of information (sample size, methodology, results, limitations), and organizes everything into a table.
You can ask questions like “What interventions have been tested for reducing procrastination in college students?” and get a structured table of studies with their findings side by side. It’s not perfect no tool is, but it cuts the screening phase of a literature review from days to hours.
The key is asking specific questions. “Studies on learning” will return garbage. “RCTs testing retrieval practice versus re-reading on exam scores in undergraduates” will return a focused, usable list.
For a thesis student or anyone doing a systematic review, this alone is worth learning in the first week of the semester.
5. A Local Coding Assistant (Ollama + Continue) For Programming Assignments Without the Privacy Risk
Here’s the uncomfortable reality about using ChatGPT for programming coursework: at many universities, submitting AI-generated code without disclosure is an academic integrity violation. Even if you’re using it to learn and not to cheat, pasting your assignment into a cloud model is a gray area at best.
The solution is running a model locally.
Ollama lets you run open-source LLMs (like Llama 3, Mistral, or CodeGemma) entirely on your own machine. Nothing leaves your computer. Continue is a VS Code extension that connects to your local Ollama model and gives you an in-editor AI assistant autocomplete, inline chat, and code explanations, all locally.
# Install Ollama, then pull a coding model
ollama pull codellama
# In VS Code, install Continue extension,
# then point it to your local Ollama instance in settings
The setup takes about twenty minutes. After that, you have a coding assistant that explains what your code is doing, suggests fixes, and helps you understand error messages without sending your assignment to anyone’s server.
The model isn’t as capable as GPT-4. But for understanding stack traces, writing helper functions, and getting unstuck at 1am, it’s more than enough. And it runs offline, which matters when you’re on campus Wi-Fi that blocks half the internet.
What’s the Pattern Here?
If you look at these five tools, they all solve the same category of problem: they eliminate the low-value friction between you and the actual learning.
You still have to read the literature NotebookLM just helps you navigate it faster. You still have to understand the concepts Anki just ensures they stick. You still have to write the code, the local assistant just helps you debug it.
The mistake is thinking AI tools replace the thinking. The ones worth using don’t do that. They make the thinking easier to start, and harder to procrastinate.
Good tools start with the question, “What is slowing me down?” These five tools each answer that question honestly.
Start with one. Build the habit. Then add the next.
Drop your questions in the comments.
My Code Diary publishes practical guides for programmers who want to learn faster and build smarter.
