Write an AI research paper.
1-on-1 with me - a question, clean experiments, and a paper with your name on it.
Write on the topic you care about
Examples
You'll have papers like these — and the setup to keep going
The paper is only half of it. You also walk away with a working research setup and a repeatable structure — so you can pick your next question and run it solo.
MiniMax Sparse Attention: A Small MacBook-Scale Research Note
A complete mini-paper built entirely at MacBook scale — defining a sparse-attention research program, running the experiments, and writing up the result. Exactly the kind of paper you'll produce.
Let Attention Control Its Sharpness: A Per-Head Query-Gain Scalar
One learnable scalar per attention head, zero-initialized so the model starts as the exact baseline — and it improves validation loss by −0.078 for just 144 parameters. A clean single-mechanism result.
Cheap Multiplicative Gates: the value-channel gate that survived a 10× scale-up
Screened twelve gating mechanisms, kept the one that survived everything: a per-channel value gate that improves val loss by −0.1075 and grows with scale — reported alongside eleven honest failures.
Analysis and Design of Novel Optimizers for Neural Networks
A complete thesis comparing the Muon and Adam optimizers for training LLMs and neural networks — motivation, method, experiments, and results. The upper bound on how far this work can go.
What you walk away with
Not a certificate — proof you can do research.
The 8-class path
Eight classes. One real paper.
60-minute 1-on-1 classes, personalized to your topic. Class 1 is the standalone $40 Discovery Class.
Onboarding, topic & repo setup
Pick your track — LLMs, voice, video, or RL — and set up your repo and environment so a baseline is ready to train. This is the standalone $40 Discovery Class.
Literature review & first experiments
Read the handful of papers that actually matter for your track, pick the techniques worth testing, and run your first baseline to set the number to beat.
Pick one change & go deep
Commit to ONE change versus the baseline — the core of your paper — and understand it deeply enough to implement and defend it.
Experiment design & ablation
Design the minimal ablation that isolates your change — the controls, the seeds, and the single metric that decides the answer.
Run, debug & measure
Launch the runs, catch the silent failure modes that quietly void experiments, and confirm your numbers are actually trustworthy.
Read the evidence & go deeper
Interpret signal vs noise and what the ablation proves. If it's interesting, design one follow-up that explains WHY it works or fails.
Write the paper
Write it the way researchers do — results first, then method, abstract, and charts — turning your experiment into a clear, short paper.
Polish & publish
Final edits, a clean GitHub repo and README, and shipping it publicly — with your name on it and the skill to run the next one solo.
1-on-1 · runs on a single GPU · all code & charts provided.
Full curriculum with examplesPricing
Three ways to start — founding price
Try a single $40 class, ship a mini-paper in 90 minutes, or commit to the full 8-class path and write your own. I'm keeping the first group small so every paper gets real 1-on-1 attention.
1-on-1 Discovery Class
Single 60-min 1-on-1 class
One 60-minute, one-on-one class to find your research question and map the path to a paper. The lowest-risk way to start — and it's Class 1 of the full path.
90-Min Mini-Paper
One 90-min 1-on-1 · automated AI research
I prep the topic and setup; a swarm of agents runs 30–40 experiments live, finds a result, and you leave with a real mini-paper — plus the pipeline to do it again. The fastest way to a paper.
8-Class Paper Bundle
Eight 60-min 1-on-1 classes · personalized path
The full thing: eight 60-minute 1-on-1 classes that take you from picking a topic to a short, real paper, on a path personalized to you (full curriculum below).
Keep going · monthly
Or join the community and write results every month
Write real research results alongside the group
A monthly membership where you run the rolling mini-paper challenges, post your real reproducible results, and get them read and critiqued — so you keep shipping papers long after the 1-on-1 work is done.
Cancel anytime · join in minutes
Work 1-on-1 with Vuk Rosić
Founder, Open Superintelligence Lab — I publish LLM research in public
I run an open AI research lab and a YouTube channel where I build language models, custom optimizers, and training pipelines from the ground up — and publish the papers, wins and null results alike.
Working together, I'll guide your experiment the same way I run my own: a sharp question, a clean baseline, honest measurement, and a paper at the end.
Frequently asked questions
What exactly do I walk away with?
A short, finished research paper with your name on it — one focused question, clean experiments, abstract, method, results, and charts — plus a public GitHub repo. A real, reproducible paper like the ones above, the kind a working researcher actually ships.
What's the difference between the two offers?
The $40 Discovery Class is a single 1-on-1 session to pick your topic and map the path. The 8-Class Paper Bundle is the full personalized path that takes you all the way to a short, finished paper.
Is this a cohort or a course?
Neither. Every class is one-on-one and scheduled with you directly. There's no group cohort and no fixed pace — your 8-class path is personalized to your question and your level.
How long does it take?
We can talk about it, usually 4–6 weeks for all eight 60-minute classes. They're scheduled with you directly, so the pace flexes around your week — faster if you want to push, slower if life gets busy.
Do I need to be an expert already?
No. We start from a clean baseline and a question small enough to finish.
Do I need an expensive GPU?
No. Every experiment is designed to run on a single modest GPU — a free Colab T4 or a cheap cloud instance (even $0.05 / hour) is enough, including the final paper run.
What if my experiments fails?
We have enough time to find what works.
Why is it so cheap right now?
This is the founding price, and it ends at 8 founding spots or Aug 1 — whichever comes first (currently 4 spots left). I'm keeping the first group small so every paper gets real 1-on-1 attention. After that, the price goes back up.
4 of 8 founding spots left
Start with a $40 class or claim a full 8-class spot — and write a short, real paper. Founding price ends at 8 founding spots or Aug 1 — whichever comes first.
Claim a spot