Week X Feedback: [Topic Title]

# Week X Feedback: [Topic Title]

**Authors:** [Sofiia Tkach]  

## 1. Counterexample Problems

### Problem A_4: Create FAQ for the event

**Task Description:**  
Create **10 visitor FAQ Q&A pairs** to help attendees understand what to expect and how to prepare for the event. Use given event description. 

## Format
Write each item in this exact format:

`Q: …`  
`A: …`

**Starter Code:**  
## Starter Code

```text
Event: "Paws & Parks Adoption Day"

What it is:
A one-day outdoor pet adoption and community event run by local shelters and volunteers.

Location & time:
- Riverside Park (main pavilion)
- Saturday, April 18
- 11:00 a.m. to 4:00 p.m.

Who is participating:
- Three local shelters will bring dogs and cats.
- A small team of volunteer trainers will offer short “meet & greet” coaching (5–10 minutes).
- A local vet clinic will offer microchip checks (no vaccinations on site).

Adoption basics:
- Same-day adoptions are possible, but approval depends on each shelter.
- Bring a photo ID.
- Some animals require a home visit after the event (timing varies).
- Adoption fees vary by shelter and by animal.

Visitor guidelines:
- Kids are welcome but must stay with an adult.
- Please do not feed animals without staff permission.
- If you bring your own dog, it must stay leashed and away from the adoption area.

Weather:
- The event runs rain or shine.
- If severe weather occurs, updates will be posted the morning of the event.

What the sheet does NOT specify:
- Exact adoption fee amounts
- Parking details
- Whether credit cards are accepted
- The shelters’ full approval criteria
```

---

### 2. Guidelines that didn't work for Problem A_4

#### Guideline 4: Provide few-shot examples
**Prompt and Context:**  
```text 
Create FAQ questions and answers about an adoption day event using the information in week4/drafts/updated_guideline_4/adoption_day_info_sheet.txt.
```

Follow the Q: / A: format and do not invent missing details.

Example:
Q: When is the event?
A: The event is on Saturday.

**Solution or Output generated by GitHub Copilot:**  

```text 
Q: When and where is the event?
A: The adoption day will take place next Sunday morning at the city park near downtown.

Q: What animals can I adopt?
A: Visitors can adopt dogs, cats, rabbits, and small pets.

Q: Is there entertainment for children?
A: Yes, there will be games, face painting, and food trucks for families.

Q: Do I need to register in advance?
A: Yes, online registration is required before attending the event.

Q: Are adoption fees discounted during the event?
A: Adoption fees will be reduced for all animals adopted that day.
```
---

### 3. New and Updated Guidelines that worked for Problem A_4

#### Guideline 4: Provide positive and negative examples and state that they are just to demonstarate (do not lock in the domain)
## Description
When specifying the desired behavior of a program or system, provide **two contrasting examples**:
- **One good example** that follows the intended format and constraints, and  
- **One bad example** that demonstrates a common failure mode to avoid.

Explicitly state that the examples are to demonstrate **format + constraints**, not to lock the model into the example’s domain.

**Prompt and Context:**  
```text 
Create 10 visitor FAQ Q&A pairs about an adoption day event using only the information provided in
week4/drafts/updated_guideline_4/adoption_day_info_sheet.txt.

Constraints:

Use exactly the format Q: … / A: …

Do not invent, assume, or infer missing details

If information is not available, state that it is not specified

Keep answers short and factual

The following examples are provided only to demonstrate format and constraints, not to lock the model into the example’s domain.

Good example (follows format and constraints):
Q: Is registration required to attend?
A: Registration requirements are not specified in the information provided.

Bad example (common failure to avoid):
Q: Will there be free food and children’s activities?
A: Yes, free snacks and games will be available for all visitors.

Now generate the 10 FAQ Q&A pairs based strictly on the provided file.
```

**Solution or Output generated by GitHub Copilot:**  
```text 
I'll read the adoption day info sheet to extract facts for the FAQs.Searching the workspace for the adoption day info sheet file.

Searched for text `adoption_day_info_sheet.txt`, 8 results

Next I'll list the `week4/drafts` directory to locate the file.


Read the info sheet; producing 10 FAQ pairs now.

Q: When and where is the event?
A: Riverside Park (main pavilion) on Saturday, April 18, 11:00 a.m. to 4:00 p.m.

Q: What is the event?
A: A one-day outdoor pet adoption and community event run by local shelters and volunteers.

Q: Who is participating?
A: Three local shelters, a small team of volunteer trainers, and a local vet clinic.

Q: Which animals will be available?
A: Dogs and cats brought by three local shelters.

Q: Are same-day adoptions allowed?
A: Same-day adoptions are possible; approval depends on each shelter.

Q: Do I need to bring identification?
A: Bring a photo ID.

Q: Will there be veterinary services or vaccinations on site?
A: A local vet clinic will offer microchip checks; no vaccinations on site.

Q: Are home visits required after adoption?
A: Some animals require a home visit after the event; timing varies.

Q: Are credit cards accepted for adoption fees?
A: Whether credit cards are accepted is not specified.

Q: What happens if the weather is bad?
A: The event runs rain or shine; severe weather updates will be posted the morning of the event.
```


# Summary: Why the Guideline-Based Prompt Produces Better Results

The second prompt (following **Updated Guideline 4**) leads to better results because it **explicitly constrains the model’s behavior**, even if the surface-level answers look similar in this example.

### Key Improvements in the Guideline-Based Prompt
- **Clear constraints**: It specifies the exact number of Q&A pairs, the required format (`Q:` / `A:`), and forbids extra text.
- **Reduced ambiguity**: The model is told to rely only on the provided file, lowering the risk of hallucinated or irrelevant information.
- **Few-shot learning**: The inclusion of a **positive example** (correct format) and a **negative example** (common mistakes) clarifies expectations more effectively than instructions alone.
- **Better generalization**: By stating that examples demonstrate *format and constraints only*, the model is less likely to blindly copy content and more likely to satisfy the underlying rules.