Evaluating AI responses in English (India)
Data Annotation TechTextFine TuningEvaluation Rating
Project hours completed: ~165
Scope:
- Training AI bots in Indian English by providing locale-specific text, audio and image prompts; evaluating and rating bot responses on various criteria such as truthfulness, harmlessness, instruction following, etc.
- Generating model failures and providing golden responses
- Reviewing and rating the work of other linguists
The main requirement across all tasks was to use English as written or spoken in India. This makes for a tremendous learning experience for both the bots as well as the humans training them, as Indian English IMHO is not a single language but a spectrum, with highly nuanced characteristics from region to region.
Project hours completed: ~165 Scope: - Training AI bots in Indian English by providing locale-specific text, audio and image prompts; evaluating and rating bot responses on various criteria such as truthfulness, harmlessness, instruction following, etc. - Generating model failures and providing golden responses - Reviewing and rating the work of other linguists The main requirement across all tasks was to use English as written or spoken in India. This makes for a tremendous learning experience for both the bots as well as the humans training them, as Indian English IMHO is not a single language but a spectrum, with highly nuanced characteristics from region to region.
2024