Start with a question
The learning studio teaches mathematical reading, information and partitions, optimization, AI build architecture, and research reasoning. It is grounded in the Atlas’s recovered manuscripts, source code, and research notes.
Each lesson presents an objective, an explanation, a worked example, four questions with hints and solutions, and links into the original record. Recommended prerequisites guide your route; they never lock a lesson.
Practice records what you actually did
A lesson’s skill is checked after two distinct questions are answered correctly on their first valid submission in a practice round, without hints or a revealed answer. Repeating the same question does not add another distinct check. Hinted answers and corrections remain useful practice and are recorded as supported work.
You can start a new round to try again after learning from a mistake. This progress system records practice evidence; it does not issue credentials, certify mastery, or establish that a research system has achieved AGI.
Marking a lesson read only marks its reading complete. It does not award a correct answer.
Your progress stays in your browser
Learning progress is stored locally for this Site. It can survive a page reload, but clearing browser data or ending a private browsing session may remove it. If storage is blocked, the interface tells you that progress is temporary.
Use Export progress to save a JSON copy. Importing a valid copy merges recognized lesson and question records. Resetting requires a second confirmation inside the page. It resets learning progress, while the separate archive reading list keeps its own records.
The labs compute the displayed model
The seven labs perform bounded, deterministic teaching calculations in the browser. Numeric outputs, charts, partition comparisons, and state trajectories respond to their actual inputs. Animations begin only when you start them; you can also step manually.
The graph-partition lab uses a small weighted normalized-cut example. Its objective is stated explicitly and differs from the manuscript’s conditional-Gaussian normalized information-partition objective. The evidence workshop uses hypothetical scenarios for practice.
These teaching activities are distinct from activating the recovered research builds. This Site currently has no verified control connection to a live MEGAMIND, PANTHEON, or SEFERIM engine. Their pages explain the recovered source and evidence.
Follow every idea back to its source
Lessons are newly written teaching material. Source links point to the original Atlas records, where notation, interpretation, assumptions, and historical status remain explicit. All 2,910 mathematical records and 292 build records have a contextual entry into the learning paths. The full archive remains available through the research atlas.
Worked examples use small, transparent inputs chosen for teaching. Their calculated values are examples, not newly observed results from the original project engines.
How the learning experience was designed
The design alternates worked examples with problems, combines visual and verbal explanations, and uses retrieval questions to revisit ideas. These choices follow the Institute of Education Sciences practice guide on instruction and study. Khan Academy’s published mastery guidance informed the distinction between activity and demonstrated skill; this Atlas uses its own explicitly stated practice rules.
The Atlas is an independent project. It is not affiliated with Khan Academy.