Article
What happens when you check whether your own detector detects what it claims
leventbulut
• • 1LEVENT BULUT PRO
leventbulut
AI & ML interests
None yet
Recent Activity
published an article 3 days ago
What happens when you check whether your own detector detects what it claims updated a dataset 3 days ago
leventbulut/objective-projection published an article 11 days ago
Objective Projection v7.2: Closing the Pattern F Gap and Extending the Hard NegativesOrganizations
None yet
published an article 3 days ago
updated a
dataset 3 days ago
published an article 11 days ago
Article
Objective Projection v7.2: Closing the Pattern F Gap and Extending the Hard Negatives
leventbulut
• posted an update 12 days ago
Post
112
"Summarization Bias" — a short data-engineering note.
Write a scene so that no emotion word appears, then ask a model to
summarize it. It comes back as "the character feels anxious." The model
performed the exact move the text was built to avoid: it re-attached
the label.
This bites hardest in evaluation. An LLM-as-judge runs the same step
internally, silently re-labels what was shown, then penalizes text that
did its job. So the dataset ships a transparent, rule-based detector
instead of an LLM judge, plus hard negatives that mark the re-labeling
move as a negative.
Honest limit: one rule (atmosphere contradiction) detects at ~10% —
the boundary where flat pattern-matching runs out.
Learn More: https://huggingface.co/blog/leventbulut/summarization-bias
Dataset: leventbulut/objective-projection
Write a scene so that no emotion word appears, then ask a model to
summarize it. It comes back as "the character feels anxious." The model
performed the exact move the text was built to avoid: it re-attached
the label.
This bites hardest in evaluation. An LLM-as-judge runs the same step
internally, silently re-labels what was shown, then penalizes text that
did its job. So the dataset ships a transparent, rule-based detector
instead of an LLM judge, plus hard negatives that mark the re-labeling
move as a negative.
Honest limit: one rule (atmosphere contradiction) detects at ~10% —
the boundary where flat pattern-matching runs out.
Learn More: https://huggingface.co/blog/leventbulut/summarization-bias
Dataset: leventbulut/objective-projection
published an article 12 days ago
Article
Summarization Bias: Why Language Models Re-Label the Emotions You Tried to Hide
leventbulut
• • 2 published an article 13 days ago
Article
Objective Projection: Engineering Emotion in Text with Physical Parameters Instead of Emotion Labels
leventbulut
• [ANNOUNCEMENT] v7.1 micro-release: Hard negatives + Atmosphere Contradiction sub-patterns
#6 opened 13 days ago
by leventbulut
[ANNOUNCEMENT] 🆕 v7 release: applied_rules detection, CITATION.cff, academic papers
#5 opened 14 days ago
by leventbulut
[CASE STUDY] Standard AI Output vs. Objective Projection (OP) Methodology
#3 opened about 1 month ago
by leventbulut
[ANNOUNCEMENT] Objective Projection SFT Dataset v4: Beyond "Flowery" Prose through Narrative Physics
#2 opened about 1 month ago
by leventbulut
liked a
dataset about 1 month ago
Thanks for sharing this dataset!
🔥 1
1
#1 opened about 2 months ago
by 
sibersafari
published a
dataset 2 months ago