GuestYou're practising as a guest — no account needed, and the drills are the real ones. Your progress is kept in this browser only, so clearing site data ends it. Continue with Google to get an account that follows you between browsers; it starts fresh, and what you did as a guest stays here.
Targeted Practice Mode
Article 1
Why this app exists
Delta Drills is an intelligent tutoring system that serves the learner LeetCode-style problems to train PyTorch prerequisites specifically for the AI-safety ARENA curriculum. Given that it is an intelligent tutoring system, there are three markers that distinguish it in value from something like LeetCode or Khan Academy (or simply prompting ChatGPT or Claude as a tutor).
It is personalized, and AI-paced (as opposed to self-paced)
The app maintains a student model that estimates a student's strengths and weaknesses as they work. It then serves them the right concept at the right level of difficulty and scaffolding. Students move on from a concept when they have mastered it, rather than after they run out of time in a course where time conditions dictate keeping up with everyone else in your cohort (as with any in-person program).
It eliminates decision fatigue of choosing content, while preserving human connection
For this MVP, I target PyTorch, but these principles apply just as well to all the other concepts necessary for ARENA prerequisites (and beyond, to ARENA itself). In order to prepare for ARENA, the learner is left with prioritizing learning linear algebra, calculus, information theory, and neural networks, all before even touching the curriculum. Self-diagnosing and practicing the right concepts with the right investment becomes irrelevant if one has a tool that does all of that automatically. There are no modules to click on like one has with Khan Academy.
For the ARENA curriculum itself, one may wonder how this would align with a linear 8-week program. It doesn't. Or rather, not without adjustment. My vision is to give the option to go through the ARENA curriculum with Delta Drills driving it all. Interacting with other humans is also valuable, however, which is why I designed a path that syncs well with the in-person program. The curriculum already serves up suggested time limits for working on problems. If a learner spends 15 minutes on a problem they struggled with, they should spend time later shoring up the concepts that made it hard. Delta Drills queues that practice for them, so they keep pace with the cohort and the gaps still get closed.
It leverages the expertise reversal effect
Instruction that helps a novice can actively hurt someone further along. A full worked example is exactly what you need the first time you meet a concept; once you are competent, that same example gets in your way. So the app picks more than the concept. It also picks the level of support: a lesson or worked example when a concept is new to you, a partially worked example as you improve, and a bare problem once you are strong. Same material, support matched to where you stand.
How the app works
Three steps
You will spend nearly all of your time on the second one.
Step 1
Take the placement test
Start here. The placement test serves you at most six problems and takes twelve minutes at the outside, since each problem is capped at two minutes. Every answer updates the app's estimate of your understanding across every concept in the knowledge graph, not just the concept the question came from. That is why six problems are enough: when the test ends, Practice knows where to start you.
Step 2
Live on the Practice tab
Day to day, the Practice tab is the whole app. Open it and solve what it gives you. You do not pick topics or track what is due, and you do not decide when you are done with a concept; the app works that out from your answers. This is where you spend most of your time, and there is nothing else you need to manage.
Step 3
Then ARENA itself
Whenever you are ready, the app brings you over to the ARENA curriculum itself, landing you on the exact coding problem for the ARENA exercise you are ready for. That handoff is the point of everything before it.
Under the hood
One loop, run against your own frontier
1
A placement test places you
A short placement shows you where you actually stand against what ARENA will ask of you.
2
It picks the concept
The app chooses the next concept to teach you, matched to your current level of mastery.
3
And picks how to teach it
For that concept it serves one of three things: a lesson, a worked example, or practice problems.
4
Your answer moves the model
How you do updates the student model, so the next choice is made on newer evidence than the last one.
5
The scaffolding fades
As you solve problems the model tracks your learning frontier and pulls the support back to match it.
6
Then the real thing
Once you're ready for an ARENA problem, it gives you the ARENA problem.
The map underneath: every concept, and what it needs first
The app knows the curriculum as a graph: each box is one concept, and an arrow means you need the concept it comes from before the one it points to. That is what lets it tell "you have not learned this yet" apart from "you were never taught the thing underneath it".
ARENA is the standard on-ramp from "motivated and numerate" to "can implement a transformer and run interpretability experiments." Its exercises assume fluency in PyTorch tensor manipulation, einops, einsum, linear algebra and probability. Cohorts arrive with very different amounts of that fluency.
Our working hypothesis is that two things go wrong at the start of the ramp. Readiness is guessed, not measured: organizers learn who was under-prepared only after the course begins. And when a learner stalls, the real cause is usually upstream — an exercise on attention fails because a broadcasting rule never stuck — so the help they get (a TA, an LLM, re-reading the notebook) treats the symptom.
These are hypotheses, not findings. The next step is interviewing ARENA learners, facilitators and organizers about what actually happened in their last cohort; what we learn goes in § Evidence rather than asking anyone whether they would like the tool.
Theory of change
We still need humans who understand what the AI is doing.
Why this is safety work and not ed-tech, in three legs. None of them is shown by data yet; each names what would test it.
Leg 1 · oversight
Better and cheaper human oversight
You can't bootstrap trust in an AI that may be manipulating you. Control helps, but the last line is a human with enough technical understanding to judge what the AI says and catch it trying to deceive. Human review is expensive; a reviewer who actually understands the model's code and claims reviews better and faster. Upskilling raises the quality of oversight and lowers its cost.
Leg 2 · catch-up · Fig. 1
A bigger safety research pool, sooner
AI research effort grows ~25× a year; human research effort ~4%. We can't steepen the human curve. We can shorten the time it takes to become a useful safety researcher, which grows the pool and moves researcher-years into the window where they matter. Experts benefit too, as new sub-fields and tools appear.
Leg 3 · distribution
Built for safety, on purpose
Dual use is real: in capabilities hands the same tool speeds up capabilities. So Delta Drills is non-profit by design (Math Academy is for-profit, ~$50/month) and deliberately passes on capabilities money, and it is distributed only through safety networks: GEAR, AISC, local AI-safety groups, with a curriculum scoped to ARENA.
Leg 2, made explicit
AI research effort, and the human understanding that has to keep upFig. 1 · hypothesis · Fermi model
Why the green line matters: humans who understand the system → align the trusted model → which oversees the untrusted model doing the work. Understanding is what lets humans hold it accountable and correct it.
This is a hypothesis, not a result. After MacAskill & Moorhouse's chart (AI research effort ≈25×/yr), with AI effort split into capabilities and safety. The human side is shared human understanding: safety researchers who understand the systems well enough to check them, in researcher-equivalents. AI safety research here is humans directing AI agents, so it grows with that understanding. AGI marks the year AI research overtakes human understanding; move it to test short and long timelines. Delta Drills helps in two ways. Learners (default 1.9×) reach useful sooner, freed program budget reopens seats, and remote learners join, so the pool grows. Experts (default 1.3×) keep up with superhuman systems through a tutor the AI itself generates, so the same people understand more. Pool model: 1,000 people a year (growing 15%/yr) start a 12-month ramp; 40% finish; researchers leave at 15%/yr. Both effect sizes are estimates, not measurements; measuring them is what the pilot is for. The model shows the mechanism, not a forecast.
Safety is talent-constrained at the rampMore people want to do technical safety work than can get research-ready quickly.Assumes: oversubscribed programs; long self-study times.
The ramp stalls on prerequisitesStalls trace to specific, diagnosable gaps upstream of the exercise.Untested: interviews start this autumn.
Delta Drills closes those gapsPlacement finds the frontier; drills teach only what is missing; review keeps it.Mechanisms in § How it works.
Shorter ramp, fewer drop-outsLearners reach ARENA-ready sooner and more of them finish.Measured in the AISC pilot.
More researcher-years, earlierBanked before the period when AI research effort dwarfs ours.Fig. 1 makes this explicit.
The architecture follows Math Academy's published design — a knowledge graph, a student model, a diagnostic, and spaced repetition that understands how skills contain one another. We rebuilt it for a different domain: instead of multiple-choice mathematics, learners write PyTorch that is executed and graded, and the graph is the prerequisite structure of the ARENA curriculum. Every figure below runs the same update rules as the production backend.
Two ways through the same content
Read ARENA in order. Drills patch the gaps. The ARENA notebooks themselves are built into Delta Drills and read section by section, in order; ARENA stays the course. Under every exercise sits one button, e.g. Practice make_rays_1d: it starts a short timed block of drills on the concepts that exercise uses. A miss inside the block pulls drills on that concept's direct prerequisites to the front of the queue. When the block ends, the learner goes back to the notebook.
Linear path · ARENA notebooks, in-app
0.0 · Prerequisites
0.1 · Ray tracingexercise make_rays_1d: stuck → “Practice”
0.2 · CNNs & ResNets
0.3 · Optimization
1.x · Transformers & interpretability …
press Practiceblock done
Adaptive layer · knowledge graph + drills
Timed block on the exercise's conceptsanswer + review time and count set by the learner
A miss pulls in direct prerequisitestheir drills jump the queue; code is run and graded
Block ends → back to the notebooka paused block resumes from the same button
The knowledge graph
Each node is a concept a learner can be tested on; each arrow says "you cannot do this without that." Grey nodes are prerequisites, and nodes are coloured by the lesson they belong to. Click any concept to light up everything it depends on and everything that depends on it.
The ARENA prerequisite graphFig. 2 · live data
Loading the map…
No estimateInferredlessmore mastered
Hover a concept to light up everything it needs first. The real graph shipped in the app today: 82 concepts and 211 prerequisite edges, from Python basics up to building a ResNet in ARENA 0.2. The AISC project maps the rest of ARENA, chapter by chapter.
Placement: find the frontier, fast
A learner's knowledge frontier is the boundary between what they know and what they are ready to learn next. Testing all 82 concepts one by one would take a day. Instead, every answer moves belief across the graph: passing a concept is evidence you know its prerequisites; failing one is evidence against what depends on it. The test always asks about the concept whose answer is expected to remove the most uncertainty across the whole graph. Play the learner below.
Adaptive placement on the graphFig. 3 · interactive
Next probe · max expected information——
knownuncertainunknownfrontier
Belief per concept starts at 50%. A correct answer is strong evidence (Bayes factor ≈16), a wrong one weaker (≈0.21, careless slips are common in code), "I don't know" strongest of all (≈0.05) — the same asymmetry ALEKS uses. Evidence propagates up to 3 hops, halving per hop toward prerequisites. Frontier = not yet known, but every prerequisite is. The production test also caps each problem's time and draws probes from ARENA's own exercises first.
Spaced review: remember what you learned
Memory decays. Reviewing just as recall starts to slip resets the curve and makes the next decay slower, so each well-timed review buys more time than the last. Drag the review markers, or let the scheduler place them at the retention target. Learners and topics differ in speed: a concept that comes easily needs fewer reviews; a hard one needs more, sooner.
Built vs. planned. Delta Drills today decays every belief with a fixed 14-day half-life: a faded concept rises in the queue and resurfaces, but a review does not lengthen the next interval. Expanding, per-learner intervals are what Math Academy has and what the AISC team would fit from real logs. Switch between the two below.
The forgetting curve, and what reviews do to itFig. 4 · interactive · built vs. planned
review = 1 repimplicit = 0.3 rep
Drag the review markers under the plot. Each successful review multiplies the memory's half-life, more when recall had dropped further (desirable difficulty); the scheduler reviews when recall drops to 70%. Implicit reviews (thin bars, +0.3 each) are the fractional repetitions a concept gets when a harder concept that encompasses it is practised — see Fig. 5. They count toward its repetitions and lengthen its half-life, so the scheduler waits longer before the next full review. In Production today mode the half-life stays at 14 days whatever you do, so reviews come at a constant rhythm; the real engine also decays toward the prior rather than to zero. Planned mode is illustrative: the speeds and growth factor are not yet fitted to anyone. Fitting them is Stream 2.
Implicit review: one drill, several concepts
Flashcard schedulers treat every fact as independent. Skills are not: writing a reduction along an axis exercises broadcasting and aggregation. When the graph says a concept encompasses a simpler one, a correct answer on the advanced concept credits the simpler one with a fractional repetition. Due reviews can then be knocked out by new learning — the fewest pushes to topple the most dominoes.
Fractional implicit repetition on real edgesFig. 5 · interactive
Click a coloured concept.
Node fill = P(known) under per-concept Bayesian Knowledge Tracing (prior 0.10, learn 0.30, guess 0.20, slip 0.10). Edge labels are the encompassing weights authored in the curriculum: a correct answer on the upper concept adds a 0.30 × weight learning step to the lower one. Wrong answers credit nothing downstream. Forgetting regresses every belief toward the prior with a 14-day half-life.
When a learner struggles: go down the graph, not down the difficulty
The usual response to a miss is an easier problem on the same concept. That keeps the learner on the symptom. After two misses since their last unaided success, Delta Drills redirects to the weakest prerequisite that still has fresh drills, serves a dose of three, then returns. Interleaving is enforced: no concept more than three times in a row while other frontier work exists.
Remediation by prerequisiteFig. 6 · interactive
——
Target: attention-pattern einsum, ARENA's first real bottleneck. Miss it twice and the picker walks to the weakest one-hop prerequisite, then its prerequisites. A deliberate miss buys at most three easier drills per prerequisite, so the rule cannot be gamed into an easy streak.
What is new here
Same backbone, different animal.
We credit Math Academy for the architecture and borrow its best explanations. What we built on top is specific to research engineering, and several pieces are original.
Math Academy
Delta Drills
Domain
School and university mathematics
The prerequisites of ARENA: PyTorch, einops, einsum, linear algebra, probability
Answers
Numeric / multiple choice
Code, executed in a sandbox and checked against test cases; near-miss feedback for wrong dtypes, shapes and calls
Student model
Knowledge profile of spaced repetitions
Per-concept BKT with implicit credit plus an additive-logistic model (generalising IRT and Elo/Glicko) that gives P(correct) with a credible interval
Placement
Question-count diagnostic
Time-budgeted (1 h / 3 h / 6 h), value-weighted by how much of the course each concept unlocks, probes with ARENA's own exercises first, and reports prerequisite edges your answers contradict
Readiness
Mastery gates on prerequisites
Before each drill, asks whether the lower bound of P(correct) clears the bar — and whether re-reading the lesson would lift it. If so, the lesson comes back first
Relationship to a curriculum
Is the curriculum
An adaptive layer around an existing one: ARENA stays the course; we get people ready for it and unstuck inside it
We are deliberately not collecting "would you use this?" answers. Interviews ask about the last cohort: how readiness was judged, the last participant who fell behind, and what the facilitator actually did about it. The product is shown only after the problem is understood, and conversations end with a request for a concrete next step.
Weak"That's really cool." · "Students would probably benefit."
BetterAgreed to a second meeting · introduced us to a facilitator · took the placement test themselves
StrongPut incoming participants through placement · tried remediation with a stuck learner during a live cohort
Case study: one learner's September
The first learner is the project lead. Below is every graded answer, placement probe and lesson read from September on the ARENA prerequisites, pulled from the production log and scored three ways. It is one person, so it proves nothing about anyone else; what it does show is where the model is wrong, which is where Stream 2 starts.
≈48 h studied in the app since Jul 21.
September on the ARENA prerequisitesFig. 7 · real data · n = 1, the author
Hover a day.
app scoredemonstratedtried at allactual correctmodel predicted
Top: the app's score (mean per-skill posterior, prior 0.02, 14-day forgetting) against plain evidence: the share of the section's concepts whose latest graded answer was right, inside the share that has been tested at all. Middle: over the trailing 7 days, how often answers were right versus the engine's own prediction for those same drills — . Bottom: study time per day, estimated from event timestamps (placement problems are timed exactly; other events credit the time since the previous event, capped at 20 min). Lesson re-reads and time spent reading ARENA notebooks are not logged, so these hours are a lower bound. Daily aggregates only; no answers or item ids are published.
What this shows. The engine under-predicts this learner on the drills it serves (the violet line sits above the blue one all month). But most of the gap between "feels further along" and the score is concepts nobody has asked about yet: in 0.0, 7 of 13 have no graded answer at all, and no model can credit those. Reading a lesson changes nothing in the score because only the most recent read per concept is stored — so whether reading teaches cannot be tested from this log. Logging every read is the first Stream 2 fix.
Where things stand, honestly
Built
Live app; 14 lessons, 82 concepts, 1,585 drills; engine, placement and remediation; ARENA notebooks in-app.
Users
EAGx Berkeley: 7–8 people tried it when it wasn't ready (weak content, confusing UX). Since the rebuild, one outside learner, and he chose to do ARENA in Delta Drills instead of Google Colab. That is a commitment, not a compliment, but he is a friend: n = 1, biased. The only sustained user is the author (above).
Interviews
None formal yet. Next: the local AI-safety group, then the GEAR online ARENA cohort (Oct–Dec, applying now), where the author works through ARENA inside Delta Drills alongside other learners. Questions are about their last stuck moment, not about the product.
The crux. Do AI-safety learners actually want this, beyond saying they're interested? Nothing on this page answers that yet. It is the first thing the next month is for.
Metrics by job
Placement
Completion of the diagnostic; whether it separates learners meaningfully; whether recommendations change what they study; whether organizers use the result.
Remediation
Triggered at meaningful moments; completed; learner returns to the ARENA exercise and progresses; less facilitator time; voluntary re-use.
Practice
Return rate without reminders; where learners stop; calibration of predicted vs. actual correctness.
Delta Drills is an adaptive tutor for the prerequisites of AI safety research engineering. It maps what a learner already knows onto a knowledge graph of the ARENA curriculum, teaches only what is missing, and keeps it from decaying.
Working product · user validation just starting (n=1 so far)
● Built
82 concepts on a prerequisite graph, 1,585 executable drills, a graph-wide placement test, spaced review, and remediation that walks down the graph.
● Being tested now
Do ARENA organizers discover prerequisite gaps only after a cohort starts? When learners stall mid-course, is the cause a missing prerequisite?
● Not yet known
Whether a shorter ramp translates into more safety research, and whether organizers will change their workflow to use it. The AISC project is designed to find out.
Project plan & fit for AISC
Map all of ARENA, learning it as we go.
The software is mostly done; the remaining work is content. Deliverable: the whole ARENA curriculum mapped, as a knowledge graph plus verified drills, by the end of April.
The team is the first cohort. Every teammate works through ARENA inside Delta Drills as a learner and reports where the content or the graph is wrong; the team revises both as it goes. Content is written just in time, one chapter ahead, because ARENA's material already exists and each later chapter gets authored with a stronger model. Teammates aren't neutral users, so an outside group still matters for evidence.
NowAuthor in the appARENA ch. 0 mapped; own learning logged.
Sep 25GEAR applicationOnline ARENA cohort.
Oct–DecGEAR pilot, if acceptedThe author and the first outside learner (both applying) do GEAR's ARENA inside Delta Drills: a 2-learner pilot, 13 weeks at 5 h/wk, plus cohort feedback.
Jan–Apr 2027AISCThe team maps the rest of ARENA.
Stream 1 · content
The rest of ARENA
Chapter by chapter, just in time: concepts, prerequisite edges, encompassing weights, and drills that pass the existing content gates. Seth is the main author; an ML-strong teammate reviews and cleans what he generates.
Stream 2 · learner model
Learning theory, fitted
Seth leads. Fit BKT and logistic-engine parameters to the team's own attempt logs; publish calibration (Brier score, reliability curves); replace the fixed half-life with fitted memory stability; log every lesson read.
Stream 3 · pilot & evaluation
Does the ramp get shorter?
Teammates' learning as the first data, plus an outside group (GEAR cohort, local AI-safety group). A quantitative A/B comparison of learning outcomes: ARENA with Delta Drills vs. the usual way (notebooks in Colab), with the same checkpoint tests given to both arms during the course and at the end; observe, don't coach.
Jan 9–10
Opening weekendOnboarding into the app, the codebase and the content gates; everyone starts ARENA in-app.
Weeks 1–15
LearnTeammates work through ARENA in Delta Drills as real learners.
Weeks 2–15
ReviseTheir feedback fixes drills and prerequisite links; the next chapter is mapped one step ahead of the team.
Weeks 3–12
Learning theoryKnowledge graph, spacing and mastery model fitted to the team's attempt logs.
Weeks 13–15
Write-upWhat changed and what didn't, with open data and negative results, on LessWrong / Alignment Forum.
Apr 23–26
Final presentationsThe mapped curriculum, calibration and A/B results, and a decision: scale, pivot, or stop. Stop condition: Colab vs. Delta Drills, same checkpoint tests; stop if there is no learning gain. A learner going back to Colab is feedback, not failure.
Risk to the runway. If GEAR rejects the application, the Oct–Dec cohort-feedback channel is gone; the fallback is the local AI-safety group and ARENA alumni recruited directly.
Uncertainties & downside risk
What could go wrong, and what we do about it.
Key uncertainties, as questions
Users
Do AI-safety learners actually want this?
Beyond saying they're interested. Today: 7–8 EAGx testers when it wasn't ready, one outside learner since (a friend, who switched from Colab to Delta Drills for ARENA), no formal interviews; the author is the only sustained user. How we find out: the GEAR cohort (if accepted) and the local AI-safety group, asked about their last stuck moment.
Timelines
Is upskilling too slow if AGI comes soon?
Maybe it is better to deploy the skills people already have. Our view: short timelines are why fast upskilling matters. The value is near zero if the window is months, and highest at 2–5 years. This is an open crux, not a settled answer.
Also open
Mechanism and measurement
Is a missing prerequisite really what stalls people? Untested. Does the engine measure right? On the one learner with data it under-predicts by 13 points (Fig. 7).
Downside risks
Capabilities
Teaching ML skills helps everyone
Non-profit by design, so there is no capabilities revenue to chase; distributed only through safety networks (GEAR, AISC, local groups); the graph is scoped to the safety curriculum (ARENA). Capabilities labs are not ramp-constrained; the safety field is. No novel ML research is produced.
Miscalibration
False confidence in a placement
Every estimate carries an interval; low-evidence concepts show wide bands by construction. Placement output is advice to a learner, never an admissions filter without a human in the loop.
Goodhart
Drilling syntax, not research taste
We measure success downstream — ARENA exercises completed, facilitator time saved — not drill streaks. Delta Drills is the prerequisite layer, not a replacement for doing research.
Opportunity cost
Organizers' time
Pilots are small and opt-in; we do the analysis. Colab vs. Delta Drills, same checkpoint tests: if there is no learning gain, we say so publicly and stop.
Project lead
Seth Gibson
Background
Maths, software and learning science
Mathematics degree.
Software: built the whole system: live app, learner model, placement, remediation, 1,585 drills, ARENA notebooks in-app.
Learning theory: how people learn and how to learn: mastery learning, spaced and implicit review, knowledge tracing, adaptive placement.
Dogfooding: learning ARENA inside the app himself: 48.5 h since Jul 21 (29.5 h in September), 150 drills answered, a lower bound (Fig. 7).
Not an ML researcher; an ML-strong teammate reviews the content.
Commitment
Full-time on AISC
AISC requires at least 10 h/week; the lead commits full-time.
Team experience
First project-lead role
Organized with a group at EA UT Austin; no team-project experience beyond that. This would be the first time leading a research team.
Open questions
Where is this weakest?
This page is written to be critiqued. One question per AISC criterion:
Theory of change. Is ramp-up actually what limits AISC teams, or is it something else?
Plan & fit. Is content + calibration + pilot + write-up too much for three months? What would you cut?
Downside risk. Are non-profit and safety-only distribution a convincing mitigation?
Project lead. What would make you confident I can lead a team on this?
Next month. With one month before the Nov 1 deadline, where would you put it?
Try it — no account needed. Everything works signed
out, and your progress is saved to this browser. Signing in with
Google starts an account you can use on any device; what you do as a
guest stays in this browser.
Loading practice focus…
Loading the graph…
Instructor review
You're reviewing the content, not practicing it. Everything you flag lands with a maintainer.
Tap an arrow to inspect it.
Loading the lesson graph…
Split a PDF
Upload a PDF and optionally a chapters CSV. Use Auto TOC if you want the backend to detect chapters.
Job status
Download chapters
Account
Your Delta Drills account.
Email
Not signed in
Status
Guest
ThemeLight and Dark are new; Blue is
the palette the app has always had and stays the default. Applies
immediately and is remembered in this browser.
What this app is for
Three courses, each with its own drills: ARENA, Delta Drills and LeetCode Patterns.
ARENA takes you from PyTorch fundamentals to original AI-safety
research, and it assumes you can already write the code —
the tensor manipulation, the einops, the training loops. Delta
Drills is this app's own small course: what mastery and spaced
repetition mean, and why the app drills you the way it does.
LeetCode Patterns is coding-interview problems grouped by the
pattern that solves them, with no lessons: you practise straight
away. This app is the practice engine for whichever course you enable below —
it drills the mechanics, tracks which ones you have, and for ARENA,
sends you to the real exercise once you're ready for it.
You're on the main edition — drills are solved in the app's own code editor.You're on the Colab edition — drills open as Colab notebooks, and the app tutors alongside them.
Groups
Follow everyone's goals, daily practice, and competency over time. Start a study group, join a listed one, or paste an invite link.
Concept Chat
—
ready for the ARENA curriculum
Each problem has its own clock, set by its concept; a LeetCode problem's is set by its difficulty and pattern.
What you are learning
Each bar is this area's estimated readiness. Take the placement test from Account and Settings to measure them instead of assuming them.
How much you practiced this week
Problems answered each day, Monday through Sunday.
0Answering--:--
Feedback on this lesson
About this lesson
Imported helpers
Target Image
Loading image preview...
🔦 This is a PyTorch drill — run it in Colab
The in-browser runner can't import torch (sign in to grade PyTorch drills right here), so this one runs in a Colab notebook. Open it (it's the problem — starter code, no answer), work through it there, then tell us how it went. Stuck? Show solution opens the worked answer separately.