Learning path

AI researcher

Read papers, run experiments and work on interpretability and alignment.

26 lessons · 172 min left

Start path → Reading a paper in 30 minutes

How to do AI researchLesson 324: Reading a paper in 30 minutesYou do not need to read every word of a paper to understand it. Read it in three passes, take notes in a fixed template, and learn to spot weak evaluation.Middle8 min read🧪 Lab📝 TestUp next
How to do AI researchLesson 325: Reproducing a resultThe best way to truly understand a paper is to make its result appear on your own computer. Start small, match one number, and log every difference.Middle8 min read🧪 Lab📝 Test
How to do AI researchLesson 326: Designing a clean experimentA good experiment answers one clear question fairly. Learn hypotheses, baselines, ablations, random seeds and clean data splits, with a small Python example.Middle9 min read🧪 Lab📝 Test
How to do AI researchLesson 327: Compute on a budgetYou can do real research without a big GPU cluster. Start small, use free and academic resources wisely, write efficient code, and know when you truly need more.Middle7 min read🧪 Lab📝 Test
How to do AI researchLesson 328: Writing and sharing your resultsResearch that nobody can read or check does not help anyone. Learn the structure of a paper, clear figures, honest limitations, and how to share through preprints, code, blogs and workshops.Middle9 min read🧪 Lab📝 Test
How to do AI researchLesson 329: AI research paths in IndiaThere is more than one road into AI research. Compare the main paths, learn how to find and approach mentors, and understand the rules of research integrity.Middle7 min read🧪 Lab📝 Test
Build an LLMLesson 41: Step 5 · A complete tiny GPTEvery piece assembled into a model you can actually train tonight. Under 100 lines, and structurally identical to a frontier model.Middle7 min read🧪 Lab📝 Test
Inside the model: interpretabilityLesson 309: Why look inside a modelTesting what a model does tells you a lot, but not why it does it. Interpretability tries to read the mechanism itself.Advanced6 min read🧪 Lab📝 Test
Inside the model: interpretabilityLesson 310: Probing: what does a layer know?Train a simple classifier on a layer's activations to see what information is there. Then learn why 'it is there' is not the same as 'the model uses it'.Advanced5 min read🧪 Lab📝 Test
Inside the model: interpretabilityLesson 311: Attention maps and their limitsAttention weights show where a head looks. They do not show what it takes from there, or what the model does with it.Advanced6 min read🧪 Lab📝 Test
Inside the model: interpretabilityLesson 312: Features and superpositionOne neuron often responds to many unrelated things. A leading explanation is that models pack more features than they have dimensions.Advanced6 min read🧪 Lab📝 Test
Inside the model: interpretabilityLesson 313: Sparse autoencodersLearn a large dictionary of sparse features from a model's activations, so that each activation is explained by a few readable pieces.Advanced6 min read🧪 Lab📝 Test
Inside the model: interpretabilityLesson 314: Circuits: tracing a behaviourFind which parts of a model cause a behaviour by swapping activations between runs. Induction heads are the classic worked example.Advanced6 min read🧪 Lab📝 Test
Inside the model: interpretabilityLesson 315: Steering models with activationsAdd a chosen direction to a model's activations while it runs, and its behaviour shifts. Useful for research, and a reminder of how fragile safety training can be.Advanced6 min read🧪 Lab📝 Test
Inside the model: interpretabilityLesson 316: Interpretability for safety: promise and limitsMany hope that looking inside models will one day catch deception or hidden goals. Here is what is real today, what is a hope, and how it fits with evaluations.Advanced6 min read🧪 Lab📝 Test
AGILesson 31: The alignment problemGetting a system to do what you meant, not what you literally said. Easy with a calculator. Extremely hard with something that optimises.Middle5 min read🧪 Lab📝 Test
Frontier alignment researchLesson 317: From concepts to research methodsYou know the worries: loopholes, power-seeking, looking good in the test. Research starts when a worry becomes a question that an experiment or an argument can answer.Advanced7 min read🧪 Lab📝 Test
Frontier alignment researchLesson 318: Scalable oversightHow do you check the work of a system that may know more than you? Scalable oversight is the family of ideas for supervising AI when human judgement alone is not enough.Advanced7 min read🧪 Lab📝 Test
Frontier alignment researchLesson 319: Constitutional and rule-based trainingInstead of rating millions of answers by hand, write down the principles and let AI feedback apply them. Learn how Constitutional AI works, why it helps, and where it falls short.Advanced7 min read🧪 Lab📝 Test
Frontier alignment researchLesson 320: Debate and weak-to-strong generalisationTwo research ideas for supervising stronger systems: let AIs argue while a human judges, and study whether a weak supervisor can train a stronger model well. Learn what these experiments can and cannot show.Advanced6 min read🧪 Lab📝 Test
Frontier alignment researchLesson 321: Dangerous capability evaluationsBefore a powerful model is released, teams test whether it can do things that could cause serious harm. Learn how these evaluations are designed, why getting the most out of the model matters, and how thresholds link results to decisions.Advanced7 min read🧪 Lab📝 Test
Frontier alignment researchLesson 322: Model organisms of misalignmentBiologists study diseases in fruit flies and mice. Alignment researchers deliberately build small, controlled examples of misaligned behaviour so they can study it safely before it appears in the wild.Advanced6 min read🧪 Lab📝 Test
Frontier alignment researchLesson 323: Choosing an alignment research problemGood research starts with a good problem. Learn how to judge taste, tractability and feedback loops, find and read papers, work with little compute, collaborate, and think about the safety of publishing.Advanced7 min read🧪 Lab📝 Test
Evaluating AI systemsLesson 217: Benchmarks and their limitsWhat popular benchmarks such as MMLU, HumanEval, SWE-bench and LMArena actually measure, why they wear out, and why a high score does not mean the model can do your job.Middle6 min read🧪 Lab📝 Test
Evaluating AI systemsLesson 218: Contamination: when the test leakedModels learn from huge amounts of web text, which can include the very test questions used to judge them. Learn the signs, and how held-out and fresh test sets protect you.Middle6 min read🧪 Lab📝 Test
Evaluating AI systemsLesson 219: What to measure for your use caseAccuracy is only one number. Task success, faithfulness, safety, speed, cost and user satisfaction all matter. Decide what "good" means before you build.Middle6 min read🧪 Lab📝 Test