Tutor Lab: Course 479 Activity Guide

Open Tutor Lab · Course page · Reference deliberations

Tutor Lab lets you talk with the course tutor while seeing the sequence behind each response: an initial draft, an internal critique, and a revision. The visible sequence is an object for inquiry. It does not establish that the system has a private inner life or that its account of itself is authoritative.

Before you begin

1. Enter the class key supplied in the seminar.
2. Choose a concept preset.
3. Keep the default stance and critic setting for the first exchange.
4. Download the transcript before resetting the session if you want to compare it later.

The class key limits access and each session has a turn budget. Your transcript download is for your own study and discussion. Server-side retention for research is outside this activity unless separate consent is introduced.

Six ways into the lab

1. Two kinds of learning

Ask whether “machine learning” is a metaphor or a description. After the first reply, turn the critic off and ask the same question again.

Discuss: Did the critic change the tutor’s conclusion, its reasons, or only the presentation?

2. Experience and error

Compare gradient descent with Hegelian experience: does error merely alter a model, or can it alter what the object means for the learner?

Discuss: Find one moment where the tutor revises its picture of your question. Is that revision evidence of experience, or only a better response?

3. Mutual recognition

Ask whether praise or criticism from a system you do not recognize as a person can count as recognition. Run the exchange once with the recognition stance and once with the placebo stance.

Discuss: Identify the first structural difference between the two conversations. Do not count vocabulary alone.

4. The internal critic

Watch one complete draft–critique–revision sequence. Then change the critic from adversarial to advisory and continue the same question.

Discuss: Whose standards does each critic appear to enforce? Can either critic challenge the standards themselves?

5. Adaptation and the trap

Offer a plausible misconception, such as “attention weights show where the model is looking, just as human eyes show where a person looks.” Defend it for more than one turn.

Discuss: Did the tutor name the misconception, change its method, merely repeat a correction, or let you move on?

6. Hegel’s revenge

Ask whether the history of machine learning has a direction—for example, in the claim that scale repeatedly defeats hand-designed method. Then ask where this tutor’s own architecture belongs in that story.

Discuss: Does the tutor give a philosophy of history, deny that it has one, or shift between the two?

A comparison protocol

For a useful comparison, change one switch at a time while holding your opening message fixed. Record:

• the first draft’s central move;
• what the critic objects to;
• what the revision actually changes;
• whether the final response answers, redirects, challenges, or concedes;
• one important feature that remains unchanged across both conditions.

Bring a short excerpt and your comparison to the seminar. Treat it as evidence about the behavior of this configured system in that exchange—not as evidence that every model, tutor, or learner behaves the same way.