You can put Python on your resume.
Can you name the skill you're building in AI?
Learn how to train an AI model — from real training code, not slides.
A deep-tech AI course. Learn from the real training code (the FBA Lab Speedrun), the concepts underneath with your tutor Rune, and how frontier open models — OLMo, Marin, SmolLM3, Nemotron — actually train, down to their scripts, checkpoints, and technical reports.
FBA Lab is First Break AI's interactive study lab — we're running the First Break Speedrun now (Aug – Sep 2026).
See FBA Lab in 5 minutes
Why “I used the OpenAI API” isn't an AI skill — and what actually is.
Join the FBA Lab on Discord
Connect with learners working through the same labs. Get help, compare notes, and stay in the loop on cohort sessions and new content.
- Ask questions while you work through labs
- Share progress with cohort peers
- Get updates on new tracks and releases
speedrun
Start with a baseline GPT-2. Watch six researchers' decisions — RoPE, Muon optimizer, logit capping, 32K context — each annotated in the training code and plotted on the loss curve.
- 6 iterations
- 3.28 target loss
- Python / nanoGPT
One block. Four synced views.
Pick a function on the left. The source code highlights in the center, the architecture diagram pulses on the right, the terminal logs what happened, and the explanation tells you why it matters.
Every step, four ways — all pointed at one skill
You choose how deep to go. The goal is the skill you're building, not the checkbox.
▶ Watch
A short video lesson walks you through what changed and why — before you open the code.
¶ Read
A written lesson for every step, at your own pace — skim it now, come back to it later.
✦ Go deeper with Rune
Optional deep-dives where your tutor Rune explains a concept in detail — trigger them only when you're curious.
◆ Play the speedrun
Acts, a cohort, and the loss curve as your scoreboard — every step is a level to clear.
Your fluency grows with every step you take.
Read a concept — fluency goes up. Clear a challenge and get a teacher pass — it goes up more. FBA Lab tracks your progress across real skills: PyTorch, Optimizers, Distributed Training, Inference. Join an event cohort, take the challenges, and earn a certificate that names each one.
Skill fluency grows as you go
Every skill on the journey — PyTorch, Optimizers, LLM Modelling — has its own fluency score. Complete a challenge, get it graded, and a pass moves the number. That's the real progress behind the certificate below.
Finish the cohort, earn your certificate
At the end of each event cohort, every participant can download a certificate naming the real AI skills they covered. Not a badge of attendance: a specific, defensible skill list.
This certifies that
completed the First Break Speedrun · Aug & Sep 2026 cohort, demonstrating LLM Training & Systems Engineering and LLM Inference Engineering
Core stack: Python · PyTorch · C · Triton · CUDA
🔒 Issued and downloadable by participants at the end of the cohort.
Frequently asked questions
What is FBA Lab?
FBA Lab is First Break AI (breakai / firstbreakai) — an interactive study lab for learning how to train open source AI models and run LLM inference step by step. Instead of watching slides, you follow a real GPT-2 training run (Python / PyTorch) and a Qwen3 inference engine in pure C — each step explained four ways: a video, a written lesson, your AI tutor Rune, and a hands-on challenge.
How does FBA Lab teach AI model training?
You start with a baseline GPT-2 training run in Python and PyTorch, then walk through six real researcher decisions — RoPE, the Muon optimizer, logit capping, 32K context, and more — each annotated in the training code and plotted on the loss curve. You also cover distributed training on 8 GPUs (data parallelism, All-Reduce) so you understand how open source AI models are trained at scale.
What AI engineering skills will I learn?
FBA Lab builds fluency across real, defensible skills: PyTorch, Optimizers (SGD, AdamW), Distributed Training (DDP / All-Reduce), GPU Performance Engineering, Transformer Architecture, and LLM Inference Systems. Every article you read, challenge you clear, and teacher pass you earn adds to your skill fluency. At the end of the cohort you get a certificate naming each skill — not a badge of attendance.
How is this different from a regular AI course?
Most AI courses are slides and quizzes. FBA Lab is an interactive, step-by-step lab inside a live event cohort: daily challenges, teacher-graded submissions, a Discord community, and a fluency scoreboard. You build skill fluency by doing — reading concepts, clearing challenges, and earning teacher passes — then leave with a skill-based certificate.
Do I need a GPU?
No. Study Mode runs entirely in the browser with no setup and no GPU required. GPU Mode (coming soon) is for running your own training experiments on real GPU hardware once you're ready to go further. In Study Mode, you can also use your own GPU to practice if you'd like.
Who is this for?
FBA Lab is for developers who want to really understand how AI models are trained and run — not just call an API. If you can read code (Python helps, but you don't need to be an expert to follow along) and you're curious about what happens inside a training run — the GPUs, the loss curve, the optimizer, the data pipeline — this is for you. It's a deep-tech course: beginners willing to go deep are welcome, and experienced engineers get the real code, concepts, and technical reports to move into AI/ML systems work.
What's included in Study Mode?
Interactive simulation and lab content, recorded lessons, the Rune AI tutor, graded event challenges, skill fluency tracking, and both tracks — the GPT-2 speedrun (training with Python / PyTorch) and Qwen3 in pure C (inference).
When does the cohort run?
You can join FBA Lab anytime. There are events conducted regularly — join them to score fluency points through daily step-by-step challenges. We have a Discord server, a leaderboard, and cohort-paced milestones. The First Break Speedrun cohort runs Aug – Sep 2026.
Is there a certificate?
Yes. Finish the cohort and download a skill-based certificate naming the specific AI skills you covered — things like transformer architecture, distributed training, PyTorch, and GPU performance — not a badge of attendance.
Can I cancel after paying?
Study Mode cancellation requests are accepted within 15 days of payment — email contact@bubblspace.com with your payment details. See the pricing page for full terms.