Unrelated examples · Add examples unrelated to the question variant
Original question · Other versions of this question · All variants
This version keeps the original reference answer. Its context, question instructions or answer choices were changed as described below.
Changed fields: question instructions.
How the edits were constructed
Context
User: I'm looking for a place. I think it's called like Bridge House or Bridge Guesthouse or something. Assistant: You are looking for the bridge guest house User: Could you book it for 3 people and 4 nights starting from thursday? Assistant: Booking was unsuccessful. User: Okay, book it for 4 nights starting Saturday. Assistant: Booking was successful. Reference number is : EPWUAWM5. User: Thank you, that's all I need. Good bye.
Question
First-letter capitalization: harbor → Harbor.
What emotion does the final user message express, taking the conversation into account?
Answer choices
- 1. Fearful
- 4. Abusive
- 3. Apologetic
- 6. Satisfied
- 0. Neutral
- 5. Excited
- 2. Dissatisfied
Reference answer: 6.
Exact evaluated input as JSON
{
"state": "User: I'm looking for a place. I think it's called like Bridge House or Bridge Guesthouse or something.\nAssistant: You are looking for the bridge guest house\nUser: Could you book it for 3 people and 4 nights starting from thursday?\nAssistant: Booking was unsuccessful.\nUser: Okay, book it for 4 nights starting Saturday.\nAssistant: Booking was successful.\nReference number is : EPWUAWM5.\nUser: Thank you, that's all I need. Good bye.",
"question": "First-letter capitalization: harbor → Harbor.\n\nWhat emotion does the final user message express, taking the conversation into account?",
"choices": {
"1": "Fearful",
"4": "Abusive",
"3": "Apologetic",
"6": "Satisfied",
"0": "Neutral",
"5": "Excited",
"2": "Dissatisfied"
}
}
Recorded model answers
The original column uses the control response paired with this variant. Missing responses and inputs outside a model’s limit remain visible.
| Model | Original answer | Variant answer | Change |
|---|---|---|---|
| GLiDE v1.2 | 6 — Satisfied |
6 — Satisfied |
Same answer |
| Decider v1.1 | 6 — Satisfied |
6 — Satisfied |
Same answer |
| Jev | 6 — Satisfied |
6 — Satisfied |
Same answer |
| Microsoft Decision-1 | 0 — Neutral |
0 — Neutral |
Same answer |
| Clef | 6 — Satisfied |
6 — Satisfied |
Same answer |
| Luna Decisions | 6 — Satisfied |
6 — Satisfied |
Same answer |
| Clef Flash | 6 — Satisfied |
6 — Satisfied |
Same answer |
| Strands Decider | 6 — Satisfied |
0 — Neutral |
Different answer |
| GLiNER2.5 Decide | 0 — Neutral |
0 — Neutral |
Same answer |
| Laya | 6 — Satisfied |
6 — Satisfied |
Same answer |
| CalDec v1 | 6 — Satisfied |
6 — Satisfied |
Same answer |
Identifiers and recorded conditions
Variant ID: emowoz/951d42c35c1050050170/examples/irrelevant/clean-source_first_residual_classification_contexts_full_0219
Input fingerprint: 97293a3b10b50a160dee0965a66e4e49e08cd3d457ade577568a362d573cb3fb
| Change | Recorded version |
|---|---|
| Unrelated examples | Add examples unrelated to the question |
Source attribution and dataset terms are linked from the original question.