designed with the help of
Claude
and
Codex
This material is based on work supported by the AI Research Institutes program supported by the NSF and Intel Corporation under NSF award DMR-2433348. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.
Soundtrack composed with the Anticipatory Music Transformer (Thickstun, Hawthorne, Donahue & Liang, 2024) from seed motifs written for the game.
This research prototype logs anonymous gameplay interactions (no personal information) to help us improve it.
Try to get through as fast as you can!
You finished the full review file.
Unlocks all eligible models, gives unlimited points, makes every assignment row in the list jump-clickable.
The single content control (the old Experimental toggle is gone). Full version (core): the stable groups only. Everything: the whole game, all 27 groups. Alek's version: only the Loss Landscapes era. Colin's version: Regularization & Validation (8.x) plus Going Deeper (10.x). Kilian's version: the perceptron → neural net → language → transformer → pretraining ladder plus Probabilistic Foundations. Karthik's version (demo): the A/B study build with the arm PINNED to gradient descent and a title-screen switch to the dots control arm — for showing both conditions. CS3780 study: the REAL study build — same content, each fresh save randomizes 50/50 into the gradient-descent arm (Loss Landscapes) or the dots control arm (Draw the Best Line 1.1–1.10), no demo controls, with the in-game pre/post concept test gated around play.
Re-seeds all hidden units randomly and discards layer-2 and momentum so training starts fresh. Use when the model is stuck (dead ReLU, flat accuracy).
Consent screen + the in-game concept test (SPIKE). The pre/post test logs answers anonymously to the interaction log below — no email/NetID, nothing sent (endpoint off). Export the log to see the responses.
Anonymous gameplay events (chapters, model choices, purchases, overheating). Stored locally on this device; export the JSON to analyze.
Study: Learning Outcomes from an Idle Game Teaching Machine-Learning Concepts
Principal Investigator: Erik Andersen, Cornell University
Funding source: National Science Foundation
This game is an optional study aid. It is not part of this course (CS 3780): playing is voluntary, it is not graded, and it has no effect on your grade or standing in the class. You must be 18 or older to participate.
Everything happens inside the game. If you join, we ask you to:
When you play, you are randomly assigned to a specific section of the game.
Time required: the study should take about 30 minutes.
We will do our best to keep your participation in this research study confidential to the extent permitted by law; however, it is possible that other people may need to review the research records and may find out about your participation in this study. For example, the following people/groups may check and copy records about this research:
In this study, however, the research records are anonymous — they contain no identifying information, so reviewing them cannot reveal who participated. There are no other foreseeable risks beyond everyday computer use, no direct benefit to you, and no compensation. The research may improve how machine learning is taught.
The data are anonymous. They are stored on Cornell-managed storage. Results are reported only in aggregate or de-identified form, and de-identified data may be retained indefinitely and shared with other researchers.
Playing the game does not require joining the study: if you choose “No,” the game works exactly the same and sends us nothing. Choosing whether or not to participate in this study will have no effect on your academic standing, record, or relationship with Cornell or with your own institution. You may stop participating at any time simply by closing the game. Because the data are anonymous, we have no way to find your records afterward, so we cannot delete data already collected.
About the study: Erik Andersen, ela63@cornell.edu. About your rights as a research participant: Cornell IRB for Human Participants, irbhp@cornell.edu. You may also report concerns anonymously via Cornell’s Ethicspoint.
By selecting “I agree” you confirm that you are 18 or older, that you have read the information above, and that you agree to take part in this research. No signature is collected and no identifying information is recorded; your choice is stored only on your own device so the game does not ask you again.