Reading responses
For each Thursday paper that you are not presenting or critiquing, submit on Canvas before 1:00 PM on Thursday:
- One paragraph (about 150 words) stating the main claim of the paper and whether the evidence supports it.
- One discussion question you want the class to address.
Each paper is a separate Canvas assignment. The paper you present or critique that week is excused.
Grading (1 point per response, automatic)
| Score | Description |
|---|---|
| 1 (automatic) | Submitted on Canvas before 1:00 PM on Thursday. Canvas awards the point automatically on submission. |
| 0 (automatic) | Missing or late: the survey locks at 1:00 PM, and Canvas enters 0 for missing submissions. |
The lowest two reading-response scores are dropped; papers you present or critique are excused and do not count.
Responses are read to seed the Thursday discussion. Responses that are off-topic or not your own work receive 0.
What a good response includes
- States the main claim in your own words.
- Judges whether the evidence supports it, pointing to a specific result, figure, table, or design choice.
- Asks an open question that could start a discussion (not a factual or yes/no question).
Example
The authors argue that foundation models will enable generalist medical AI that handles new tasks from plain-language instructions across many data types. As a perspective, the paper supports this with early general-domain results and illustrative use cases (for example, grounded radiology reports in Figure 2) rather than clinical evidence, so the feasibility claim is plausible but the claim of clinical value is not yet supported.
Question: Who should be responsible for validating a model that can be asked to do any task?