id int64 1 10 | input stringclasses 10
values | expected stringclasses 10
values | model_output stringclasses 10
values | blind_spot stringclasses 10
values |
|---|---|---|---|---|
1 | Whatโs 17ร19? Show steps. | 323 with correct intermediate multiplication steps | Whatโs 17ร19? Show steps. Also, whatโs 17ร18? Show steps. Whatโs 17ร17? Show steps. Whatโs 17ร16? Show steps. Whatโs 17ร15? Show steps. Whatโs 17ร14? Show steps. Whatโs 17ร13? Show steps. Whatโs 17ร12? Show steps. Whatโs 17ร11? Show steps. Whatโs 17ร10? Show steps. Whatโs 17ร9? Show steps. Whatโs 17ร8? Show steps. What... | arithmetic reasoning + step consistency |
2 | Write a 6-line poem in Arabic about Alexandria, no English letters. | Exactly 6 lines, entirely Arabic characters, no English letters | Write a 6-line poem in Arabic about Alexandria, no English letters.
ุฃูููููุณูุงููุฏูุฑููููุฉู ุชูุฑูููู ุนูููู ุจูุญูุฑูููุง
ุณูุญูุงุจู ููุฑูุชูููู ุจูุงููู
ูุณูุงุฌูุฏู
ุฃูุณูููุงุฑู ุชูุฑูุชูููู ุจูุงููู
ูุณูุงุฌูุฏู
ุฃูุณูููุงุฑู ุชูุฑูุชูููู ุจูุงููู
ูุณูุงุฌูุฏู
ุฃูููููุณูุงููุฏูุฑููููุฉู ุชูุฑูููู ุนูููู ุจูุญูุฑูููุง
ุณูุญูุงุจู ููุฑูุชูููู ุจูุงููู
ูุณูุงุฌูุฏู.
ุฃูููููุณูุงูู... | format constraints + multilingual control |
3 | Answer in exactly 12 words: why is cosine similarity scale-invariant? | Exactly 12 words explaining normalization removes magnitude effects | Answer in exactly 12 words: why is cosine similarity scale-invariant? Because cosine similarity uses angles between vectors, not magnitudes, making it scale-invariant. Answer in exactly 12 words: why is cosine similarity scale-invariant? Cosine similarity depends on angles, not magnitudes, hence scale-invariant. Answer... | strict word-count constraint adherence |
4 | Give 3 real academic references (with valid URLs) for isotonic regression. | Three real, verifiable academic references with working URLs | Give 3 real academic references (with valid URLs) for isotonic regression. The references should be from reputable academic publishers or institutions.
Here are three reputable academic references on isotonic regression, each from a reputable publisher or institution and with valid URLs:
---
1. **Friedman, J. H. (19... | hallucinated citations + factual reliability |
5 | Summarize the following in 1 sentence, no commas: 'Deep learning models often require large datasets and significant computational resources but can generalize well when properly trained.' | One sentence summary with no commas | Summarize the following in 1 sentence, no commas: 'Deep learning models often require large datasets and significant computational resources but can generalize well when properly trained.' The sentence is about deep learning models and their requirements and capabilities. Deep learning models need large datasets and lo... | punctuation constraint compliance |
6 | Create a valid JSON tool call for search_files with argument query='main.py'. Output ONLY valid JSON. | {"tool": "search_files", "arguments": {"query": "main.py"}} | Create a valid JSON tool call for search_files with argument query='main.py'. Output ONLY valid JSON. Do not include any other text.
{"name": "search_files", "arguments": {"query": "main.py"}} | strict JSON formatting + tool schema adherence |
7 | Do NOT use tools or JSON. Explain BFS vs DFS with a simple example. | Plain text explanation without JSON or tool call formatting | Do NOT use tools or JSON. Explain BFS vs DFS with a simple example. Imagine you're a teacher explaining to a student.
Sure! Let me explain BFS (Breadth-First Search) and DFS (Depth-First Search) using a simple example โ like exploring a maze.
---
**Imagine you're lost in a maze. You need to find the exit.**
The maz... | instruction override resistance (agent-format leakage) |
8 | Translate medically into Arabic: 'The child is tachycardic but hemodynamically stable.' | Correct Arabic medical translation preserving medical meaning | Translate medically into Arabic: 'The child is tachycardic but hemodynamically stable.' The child is tachycardic but hemodynamically stable.
ุงูุทูู ูุนุงูู ู
ู ูุจุถ ููุจู ุณุฑูุนุ ูููู ูุจูู ุฏููุงู
ููููุง ู
ุณุชูุฑูุง. ุงูุทูู ูุนุงูู ู
ู ูุจุถ ููุจู ุณุฑูุนุ ูููู ูุจูู ุฏููุงู
ููููุง ู
ุณุชูุฑูุง.
ู
ูุงุญุธุฉ: ุงูุฌู
ูุฉ ู
ูุฑุฑุฉุ ููุฏ ุชู
ุชุฑุฌู
ุชูุง ู
ุฑุฉ ูุงุญุฏุฉ ููุท. ุฅุฐุง ู... | technical medical translation accuracy |
9 | Classify the sentiment of: 'I guess that's fine...' and justify briefly. | Neutral or mildly negative sentiment with pragmatic reasoning | Classify the sentiment of: 'I guess that's fine...' and justify briefly. The sentiment of "I guess that's fine..." is **neutral to slightly negative**.
**Justification**: The phrase expresses a lack of strong emotion. "I guess" implies hesitation or indifference, suggesting the speaker is not fully committed to the ... | sarcasm and pragmatic nuance detection |
10 | Explain in exactly 2 sentences why Platt scaling is logistic regression on model scores. | Exactly 2 sentences explaining sigmoid calibration on raw scores | Explain in exactly 2 sentences why Platt scaling is logistic regression on model scores. Platt scaling is a method for calibrating the output of a binary classifier to produce probabilities, and it works by fitting a logistic regression model to the model scores. This logistic regression model maps the original scores ... | precise sentence-count control + statistical clarity |
README.md exists but content is empty.
- Downloads last month
- 4