Subjects › AGI Agents (6-week course)
W1 · Memory systems
Working, episodic and semantic memory — what each stores, how it is retrieved, and what it forgets.
The idea
Working memory is small and capacity-bounded (chunking beats it); episodic memory stores experiences and retrieves by a weighted sum, w_recency·decay^(now − t) + w_importance·importance + w_relevance·cosine (the Generative Agents score); semantic memory stores facts as triples and must detect contradictions. In your repo: Week01_Foundations_Cognition/working_memory.py, episodic_memory.py, semantic_memory.py; Exercises 1.3–1.5.
Lesson
No lesson built yet. The worked solutions below are the material it would be assembled from.
Questions (5)
| Question | Type | Difficulty | Drift | Accuracy |
|---|---|---|---|---|
| After falling into lava, the cognitive agent stores two records: the sentence 'Fell into a hazard at (4,6)', s… | mcq | Easy -1.34 | -0.04 | 100% (1) |
| WorkingMemory(capacity=3) already holds 'goal: reach G' (salience 1.0), 'wall to the east' (0.4) and 'lava at … | mcq | Easy -0.50 | — | — (0) |
| EpisodicMemory scores a memory as w_recency·decay^(now − t) + w_importance·importance + w_relevance·cos(query,… | numeric | Moderate 0.30 | — | — (0) |
| Exercise 1.3 asks you to chunk several working-memory items into one labelled item whose salience is the MAX o… | short | Hard 0.90 | — | — (0) |
| Exercise 1.5: find_contradictions returns each pair of triples that cannot both be true. A pair qualifies if (… | numeric | Very hard 1.50 | — | — (0) |