flâneur

kristinagligoric/cdi-tutorial

github.com · 632 words · saved by 1 readers

Please configure another 2FA method to reduce your risk of permanent account lockout. If you use SMS for 2FA, we strongly recommend against SMS as it is prone to fraud and delivery may be unreliable depending on your region. There was an error while loading. Please reload this page. There was an error while loading. Please reload this page. The Jupyter notebook walks you through an end-to-end, automated human-AI annotation pipeline based on Confidence-Driven Inference (CDI). We collect human annotations via Prolific or Amazon Mechanical Turk, and LLM annotations through OpenAI's API. The example is based on the method introduced in: Can Unconfident LLM Annotations Be Used for Confident Conclusions? Kristina Gligorić*, Tijana Zrnic*, Cinoo Lee*, Emmanuel Candès, and Dan Jurafsky. NAACL, 2025. https://aclanthology.org/2025.naacl-long.179/# The goal is to estimate a target statistic about a text corpus while minimizing costly human labels by: We focus on annotating texts for politeness an

Confidence-Driven Inference Tutorial What this tutorial does The Jupyter notebook walks you through an end-to-end, automated human-AI annotation pipeline based on Confidence-Driven Inference (CDI). We collect human annotations via Prolific or Amazon Mechanical Turk, and LLM annotations through OpenAI's API. The example is based on the method introduced in: Can Unconfident LLM Annotations Be Used for Confident Conclusions? Kristina Gligorić*, Tijana Zrnic*, Cinoo Lee*, Emmanuel Candès, and Dan Jurafsky. NAACL, 2025. https://aclanthology.org/2025.naacl-long.179/# The goal is to estimate a…

saved by

related reading