Retrieval
Try explaining it yourself.
Learn through an example, then try a focused recall question. Already know the topic? Start with a cold check.
Recall it coldOne nudgeExplain the gap
Understand the ML you work with.
For engineers learning a concept, closing a gap, or preparing for an interview. Explore a worked example, follow the math and code, and practise explaining it yourself.
2 free sessions, no account needed.
2full sessions$0no account164guided topics
Change the learning rate and follow the path.Study a concept
See what changes when an idea clicks.
Explore a worked correction, then answer a related question without the notes. Or open the example session to follow the full explanation.
Learn a concept
Start with a worked example. Basic programming is enough; explore unfamiliar math as you go.
Prepare for an interview
Practise explaining mechanisms and defending implementation tradeoffs.
Check what you know
Already comfortable with ML? Diagnose advanced training and systems failures.
Reinforcement learning · delayed reward
Why take a step with no reward?
A worked Q-learning example. Explore the correction, then try a different transition.
An agent moves toward a goal. The move earns zero reward, but the episode continues.
“No reward means this action was worthless. Its value should move toward zero.”
That misses what the next state makes possible. Q-learning considers estimated future rewards too.
Explore the full Q-learning session: intuition, math, and code
Q-learning: from delayed reward to an update.
An example session with math, code, and follow-up questions.
From an example to your own explanation.
Start with the idea, inspect how it works, then try using it. Ask a follow-up or request a deeper explanation at any point.
See the whole idea before the machinery.
Start with the problem, the constraints, and the mental model that makes every later equation feel inevitable.
concept
problemconstraintuse caseA record of what holds up.
Record what you can retrieve and which implementation checks pass. Two bounded queues cover six concepts: attention, position and KV cache; token loss normalization, gradient accumulation and optimizer-step correctness.
Implementation
def attention(q, k, v, mask):
scores = q @ k.transpose(-2, -1)
scores /= q.size(-1) ** 0.5
scores.masked_fill_(~mask, float("-inf"))
return scores.softmax(-1) @ v Illustrative implementation · check masking and output shape
Retention
Come back when recall is at risk.
The attention and training-step clusters use recorded gaps, hint use, elapsed time and prerequisites to suggest the next useful question.
Attention / mathdue today
Position / intuitionholding
Evidence
Not one vague mastery score.
Keep evidence by dimension, source and date. Your comparisons, model rubrics, browser checks and isolated server checks stay distinguishable.
illustrative evidence record
Systems: strong · Math: due · Implementation: 6/6 browser checks · Hint use: none
Follow the mechanism into modern ML.
Start with a concrete question. Read the primary paper, explore an available visual, or work through the topic in Study.
How reasoning models learn
Separate demonstrations, group-relative rewards, and teacher signals. Inspect a small GRPO mechanism before studying the full training pipeline.
Where inference time goes
A draft model proposes tokens; the target verifies and corrects them. Learn why the correction rule matters for preserving the target distribution.
When more thinking helps
Compare search, verification, and revision at inference time. More compute is a budget to allocate; its usefulness depends on the problem and the method.
Studio mechanisms use small illustrative models. The decoding diagram shows data flow. Neither reproduces a paper’s training runs or benchmark results.
17 authored references, open to everyone.
Deep LearningAttention MechanismsTurn compatibility scores into attention weights, then combine values. Explore the geometry, normalization, and limits of the mechanism.9 min · deep diveDeep LearningTransformer ArchitectureFollow attention, feed-forward layers, normalization, and residual connections through a Transformer block.9 min · deep diveDeep LearningBackpropagationTrace local derivatives backward through a computation and connect the chain rule to a manual implementation.9 min · deep diveLLMsKV CacheSee which token projections caching saves, what attention must still read, and how memory use grows with context.8 min · deep dive
Choose the depth. Keep the structure.
164 topics for AI-guided sessions, with 17 separately authored ML references you can read without an account. You choose the topic and pace; the tutor can explain prerequisites.
ML / Research78 topics
From classical models to training systems.
Transformers, optimization, RLHF, distributed training, serving, and the mathematics underneath modern systems.
Software engineering86 topics
From browser internals to distributed systems.
Rendering, caching, databases, data structures, backend architecture, and production tradeoffs.
Start learning. No account needed.
Only Pro takes a card. The trial does not charge for seven days. Cancel before it ends to avoid a charge.
Guestno account
$0
2 sessions total in this browser
Two complete six-stage sessions
164 topics for guided AI sessions
17 authored technical references to read
No email, no card, no profile
Sign in to save sessions and use the practice lab
Freesign in
$0
8 sessions per calendar month
8 sessions a month, no card
Saved threads you can resume
Mastery evidence and an adaptive review queue
12 retrieval evaluations a month
5 practice-lab code reviews a month
Revision cards
Optional resume context for worked examples
Pro7-day trial
$9 / month
Unlimited sessions
Unlimited review sessions
Unlimited practice-lab code reviews
Unlimited retrieval evaluations and mastery history
Everything in Free
New topics ship to Pro first
Request topics directly
A session is one topic explored across six stages. Guest sessions are a one-time allowance; Free session, retrieval, and code-review allowances reset each calendar month. Reading the authored references does not use a session.
Pick the concept you keep almost understanding.
Work through the details, ask the follow-up, and test what stayed with you.
Start a deep study session
2 sessions without an account · 8 / month when you sign in · unlimited on Pro